Geostatistical Functional Data Analysis Wiley Series in Probability and Statistics 1St Edition Jorge Mateu Editor Full Chapter

Geostatistical Functional Data Analysis
(Wiley Series in Probability and

Statistics) 1st Edition Jorge Mateu
(Editor)
Visit to download the full and correct content document:
https://ebookmass.com/product/geostatistical-functional-data-analysis-wiley-series-in-
probability-and-statistics-1st-edition-jorge-mateu-editor/
WILEY SERIES IN PROBABILITY AND STATISTICS
Established by Walter A. Shewhart and Samuel S. Wilks
The Wiley Series in Probability and Statistics is well established and
authoritative. It covers many topics of current research interest in both pure and
applied statistics and probability theory. Written by leading statisticians and
institutions, the titles span both state-of-the-art developments in the field and
classical methods.
Reflecting the wide range of current research in statistics, the series encompasses
applied, methodological and theoretical statistics, ranging from applications and
new techniques made possible by advances in computerized practice to rigorous
treatment of theoretical approaches. This series provides essential and invaluable
reading for all statisticians, whether in academia, industry, government, or
research.
A complete list of titles in this series can be found at http://www.wiley.com/go/
wsps
Edited by
Jorge Mateu
University Jaume I of Castellon
Castellon, Spain
Ramón Giraldo
National University of Colombia
Bogota, Colombia
This edition first published 2022
© 2022 John Wiley & Sons Ltd
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted,
in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, except as
permitted by law. Advice on how to obtain permission to reuse material from this title is available at
http://www.wiley.com/go/permissions.
The right of Jorge Mateu and Ramón Giraldo to be identified as the authors of the editorial material in this
work has been asserted in accordance with law.
Registered Offices
John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, USA
John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex, PO19 8SQ, UK
Editorial Office
9600 Garsington Road, Oxford, OX4 2DQ, UK
For details of our global editorial offices, customer services, and more information about Wiley products visit
us at www.wiley.com.
Wiley also publishes its books in a variety of electronic formats and by print-on-demand. Some content that
appears in standard print versions of this book may not be available in other formats.
Limit of Liability/Disclaimer of Warranty

The contents of this work are intended to further general scientific research, understanding, and discussion
only and are not intended and should not be relied upon as recommending or promoting scientific method,
diagnosis, or treatment by physicians for any particular patient. In view of ongoing research, equipment
modifications, changes in governmental regulations, and the constant flow of information relating to the use
of medicines, equipment, and devices, the reader is urged to review and evaluate the information provided in
the package insert or instructions for each medicine, equipment, or device for, among other things, any
changes in the instructions or indication of usage and for added warnings and precautions. While the
publisher and authors have used their best efforts in preparing this work, they make no representations or
warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim
all warranties, including without limitation any implied warranties of merchantability or fitness for a
particular purpose. No warranty may be created or extended by sales representatives, written sales materials
or promotional statements for this work. The fact that an organization, website, or product is referred to in
this work as a citation and/or potential source of further information does not mean that the publisher and
authors endorse the information or services the organization, website, or product may provide or
recommendations it may make. This work is sold with the understanding that the publisher is not engaged in
rendering professional services. The advice and strategies contained herein may not be suitable for your
situation. You should consult with a specialist where appropriate. Further, readers should be aware that
websites listed in this work may have changed or disappeared between when this work was written and when
it is read. Neither the publisher nor authors shall be liable for any loss of profit or any other commercial
damages, including but not limited to special, incidental, consequential, or other damages.
Library of Congress Cataloging-in-Publication Data
Names: Mateu, Jorge, editor. | Giraldo, Ramón, editor.

Title: Geostatistical functional data analysis / edited by Jorge Mateu,
Ramón Giraldo.
Description: Hoboken, NJ : Wiley, 2022. | Series: Wiley series in
probability and statistics | Includes bibliographical references and
index.
Identifiers: LCCN 2021015788 (print) | LCCN 2021015789 (ebook) | ISBN
9781119387848 (hardback) | ISBN 9781119387909 (adobe pdf) | ISBN
9781119387886 (epub)
Subjects: LCSH: Geology–Statistical methods. | Kriging. | Spatial analysis
(Statistics) | Functional analysis.
Classification: LCC QE33.2.S82 G434 2022 (print) | LCC QE33.2.S82 (ebook)
| DDC 551.072/7–dc23
LC record available at https://lccn.loc.gov/2021015788
LC ebook record available at https://lccn.loc.gov/2021015789
Cover Design: Wiley

Cover Image: © Googee/Shutterstock
Set in 9.5/12.5pt STIXTwoText by Straive, Chennai, India
10 9 8 7 6 5 4 3 2 1
v
Contents
List of Contributors xiii

Foreword xvi
1 Introduction to Geostatistical Functional Data Analysis 1

Jorge Mateu and Ramón Giraldo
1.1 Spatial Statistics 1
1.2 Spatial Geostatistics 7
1.2.1 Regionalized Variables 7
1.2.2 Random Functions 7
1.2.3 Stationarity and Intrinsic Hypothesis 9
1.3 Spatiotemporal Geostatistics 12
1.3.1 Relevant Spatiotemporal Concepts 12
1.3.2 Spatiotemporal Kriging 16
1.3.3 Spatiotemporal Covariance Models 17
1.4 Functional Data Analysis in Brief 18
References 22
Part I Mathematical and Statistical Foundations 27
2 Mathematical Foundations of Functional Kriging in Hilbert

Spaces and Riemannian Manifolds 29
Alessandra Menafoglio, Davide Pigoli, and Piercesare Secchi
2.1 Introduction 29
2.2 Definitions and Assumptions 30
2.3 Kriging Prediction in Hilbert Space: A Trace Approach 33
2.3.1 Ordinary and Universal Kriging in Hilbert Spaces 33
2.3.2 Estimating the Drift 36
2.3.3 An Example: Trace-Variogram in Sobolev Spaces 37
vi Contents
2.3.4 An Application to Nonstationary Prediction of Temperatures

Profiles 39
2.4 An Operatorial Viewpoint to Kriging 42
2.5 Kriging for Manifold-Valued Random Fields 45
2.5.1 Residual Kriging 45
2.5.2 An Application to Positive Definite Matrices 47
2.5.3 Validity of the Local Tangent Space Approximation 49
2.6 Conclusion and Further Research 53
References 53
3 Universal, Residual, and External Drift Functional

Kriging 55
Maria Franco-Villoria and Rosaria Ignaccolo
3.1 Introduction 56
3.2 Universal Kriging for Functional Data (UKFD) 56
3.3 Residual Kriging for Functional Data (ResKFD) 58
3.4 Functional Kriging with External Drift (FKED) 60
3.5 Accounting for Spatial Dependence in Drift Estimation 61
3.5.1 Drift Selection 62
3.6 Uncertainty Evaluation 62
3.7 Implementation Details in R 64
3.7.1 Example: Air Pollution Data 64
3.8 Conclusions 69
References 71
4 Extending Functional Kriging When Data Are Multivariate

Curves: Some Technical Considerations and Operational
Solutions 73
David Nerini, Claude Manté, and Pascal Monestiez
4.1 Introduction 73
4.2 Principal Component Analysis for Curves 74
4.2.1 Karhunen–Loève Decomposition 74
4.2.2 Dealing with a Sample 76
4.3 Functional Kriging in a Nutshell 78
4.3.1 Solution Based on Basis Functions 79
4.3.2 Estimation of Spatial Covariances 81
4.4 An Example with the Precipitation Observations 82
4.4.1 Fitting Variogram Model 83
4.4.2 Making Prediction 83
4.5 Functional Principal Component Kriging 85
4.6 Multivariate Kriging with Functional Data 88
Contents vii
4.6.1 Multivariate FPCA 91

4.6.2 MFPCA Displays 93
4.6.3 Multivariate Functional Principal Component Kriging 94
4.6.4 Mixing Temperature and Precipitation Curves 96
4.7 Discussion 98
4.A Appendices 100
4.A.1 Computation of the Kriging Variance 100
References 102
5 Geostatistical Analysis in Bayes Spaces: Probability

Densities and Compositional Data 104
Alessandra Menafoglio, Piercesare Secchi, and Alberto Guadagnini
5.1 Introduction and Motivations 104
5.2 Bayes Hilbert Spaces: Natural Spaces for Functional
Compositions 105
5.3 A Motivating Case Study: Particle-Size Data in Heterogeneous
Aquifers – Data Description 108
5.4 Kriging Stationary Functional Compositions 110
5.4.1 Model Description 110
5.4.2 Data Preprocessing 112
5.4.3 An Example of Application 113
5.4.4 Uncertainty Assessment 116
5.5 Analyzing Nonstationary Fields of FCs 119
5.6 Conclusions and Perspectives 123
References 124
6 Spatial Functional Data Analysis for Probability Density

Functions: Compositional Functional Data vs. Distributional
Data Approach 128
Elvira Romano, Antonio Irpino, and Jorge Mateu
6.1 FDA and SDA When Data Are Densities 130
6.1.1 Features of Density Functions as Compositional Functional Data 131
6.1.2 Features of Density Functions as Distributional Data 135
6.2 Measures of Spatial Association for Georeferenced Density
Functions 138
6.2.1 Identification of Spatial Clusters by Spatial Association Measures for
Density Functions 139
6.3 Real Data Analysis 141
6.3.1 The SDA Distributional Approach 143
6.3.2 The Compositional–Functional Approach 145
6.3.3 Discussion 147
viii Contents
6.4 Conclusion 149

Acknowledgments 151
References 151
Part II Statistical Techniques for Spatially Correlated

Functional Data 155
7 Clustering Spatial Functional Data 157

Vincent Vandewalle, Cristian Preda, and Sophie Dabo-Niang
7.1 Introduction 157
7.2 Model-Based Clustering for Spatial Functional Data 158
7.2.1 The Expectation–Maximization (EM) Algorithm 160
7.2.1.1 E Step 161
7.2.1.2 M Step 161
7.2.2 Model Selection 161
7.3 Descendant Hierarchical Classification (HC) Based on Centrality
Methods 162
7.3.1 Methodology 164
7.4 Application 165
7.4.1 Model-Based Clustering 167
7.4.2 Hierarchical Classification 169
7.5 Conclusion 171
References 172
8 Nonparametric Statistical Analysis of Spatially Distributed

Functional Data 175
Sophie Dabo-Niang, Camille Ternynck, Baba Thiam, and Anne-Françoise
Yao
8.2 Large Sample Properties 178
8.2.1 Uniform Almost Complete Convergence 180
8.3 Prediction 181
8.4 Numerical Results 184
8.4.1 Bandwidth Selection Procedure 184
8.4.2 Simulation Study 185
8.5 Conclusion 193
8.A Appendix 194
8.A.1 Some Preliminary Results for the Proofs 194
8.A.2 Proofs 196
8.A.2.1 Proof of Theorem 8.1 196
8.A.2.2 Proof of Lemma A.3 196
Contents ix

8.A.2.6 Proof of Theorem 8.2 202
References 207
9 A Nonparametric Algorithm for Spatially Dependent

Functional Data: Bagging Voronoi for Clustering,
Dimensional Reduction, and Regression 211
Valeria Vitelli, Federica Passamonti, Simone Vantini, and Piercesare Secchi
9.2 The Motivating Application 212
9.2.1 Data Preprocessing 214
9.3 The Bagging Voronoi Strategy 216
9.4 Bagging Voronoi Clustering (BVClu) 218
9.4.1 BVClu of the Telecom Data 221
9.4.1.1 Setting the BVClu Parameters 221
9.4.1.2 Results 223
9.5 Bagging Voronoi Dimensional Reduction (BVDim) 223
9.5.1 BVDim of the Telecom Data 225
9.5.1.1 Setting the BVDim Parameters 225
9.5.1.2 Results 227
9.6 Bagging Voronoi Regression (BVReg) 231
9.6.1 Covariate Information: The DUSAF Data 232
9.6.2 BVReg of the Telecom Data 234
9.6.2.1 Setting the BVReg Parameters 234
9.6.2.2 Results 235
9.7 Conclusions and Discussion 236
References 239
10 Nonparametric Inference for Spatiotemporal Data Based on

Local Null Hypothesis Testing for Functional Data 242
Alessia Pini and Simone Vantini
10.2 Methodology 244
10.2.1 Comparing Means of Two Functional Populations 244
10.2.2 Extensions 248
10.2.2.1 Multiway FANOVA 249
10.3 Data Analysis 250
10.4 Conclusion and Future Works 256
References 258
x Contents
11 Modeling Spatially Dependent Functional Data by Spatial

Regression with Differential Regularization 260
Mara S. Bernardi and Laura M. Sangalli
11.2 Spatial Regression with Differential Regularization for Geostatistical
Functional Data 264
11.2.1 A Separable Spatiotemporal Basis System 265
11.2.2 Discretization of the Penalized Sum-of-Square Error Functional 268
11.2.3 Properties of the Estimators 271
11.2.4 Model Without Covariates 273
11.2.5 An Alternative Formulation of the Model 274
11.3 Simulation Studies 274
11.4 An Illustrative Example: Study of the Waste Production in Venice
Province 278
11.4.1 The Venice Waste Dataset 278
11.4.2 Analysis of Venice Waste Data by Spatial Regression with Differential
Regularization 279
11.5 Model Extensions 282
References 283
12 Quasi-maximum Likelihood Estimators for Functional Linear

Spatial Autoregressive Models 286
Mohamed-Salem Ahmed, Laurence Broze, Sophie Dabo-Niang, and Zied
Gharbi
12.2 Model 288
12.2.1 Truncated Conditional Likelihood Method 291
12.3 Results and Assumptions 293
12.4 Numerical Experiments 298
12.4.1 Monte Carlo Simulations 298
12.4.2 Real Data Application 305
12.5 Conclusion 312
12.A Appendix 313
Proof of Proposition 12.A.1 313
Proof of Theorem 12.1 314
Proof of Lemma 12.A.2 322
References 325
Contents xi
13 Spatial Prediction and Optimal Sampling for Multivariate

Functional Random Fields 329
Martha Bohorquez, Ramón Giraldo, and Jorge Mateu
13.1 Background 329
13.1.1 Multivariate Spatial Functional Random Fields 329
13.1.2 Functional Principal Components 330
13.1.3 The Spatial Random Field of Scores 331
13.2 Functional Kriging 332
13.2.1 Ordinary Functional Kriging (OFK) 332
13.2.2 Functional Kriging Using Scalar Simple Kriging of the Scores
(FKSK ) 333
13.2.3 Functional Kriging Using Scalar Simple Cokriging of the Scores
(FKCK ) 333
13.3 Functional Cokriging 336
13.3.1 Cokriging with Two Functional Random Fields 336
13.3.2 Cokriging with P Functional Random Fields 338
13.4 Optimal Sampling Designs for Spatial Prediction of Functional
Data 340
13.4.1 Optimal Spatial Sampling for OFK 341
13.4.2 Optimal Spatial Sampling for FKSK 341
13.4.3 Optimal Spatial Sampling for FKCK 342
13.4.4 Optimal Spatial Sampling for Functional Cokriging 343
13.5 Real Data Analysis 344
13.6 Discussion and Conclusions 348
References 348
Part III Spatio–Temporal Functional Data 351
14 Spatio–temporal Functional Data Analysis 353

Gregory Bopp, John Ensley, Piotr Kokoszka, and Matthew Reimherr
14.2 Randomness Test 355
14.3 Change-Point Test 359
14.4 Separability Tests 362
14.5 Trend Tests 365
14.6 Spatio–Temporal Extremes 369
References 373
xii Contents
15 A Comparison of Spatiotemporal and Functional Kriging

Approaches 375
Johan Strandberg, Sara Sjöstedt de Luna, and Jorge Mateu
15.2 Preliminaries 376
15.3 Kriging 378
15.3.1 Functional Kriging 378
15.3.1.1 Ordinary Kriging for Functional Data 378
15.3.1.2 Pointwise Functional Kriging 380
15.3.1.3 Functional Kriging Total Model 381
15.3.2 Spatiotemporal Kriging 382
15.3.3 Evaluation of Kriging Methods 384
15.4 A Simulation Study 385
15.4.1 Separable 385
15.4.2 Non-separable 390
15.4.3 Nonstationary 391
15.5 Application: Spatial Prediction of Temperature Curves in the Maritime
Provinces of Canada 394
15.6 Concluding Remarks 400
References 400
16 From Spatiotemporal Smoothing to Functional Spatial

Regression: a Penalized Approach 403
Maria Durban, Dae-Jin Lee, María del Carmen Aguilera Morillo, and Ana
M. Aguilera
16.2 Smoothing Spatial Data via Penalized Regression 404
16.3 Penalized Smooth Mixed Models 407
16.4 P-spline Smooth ANOVA Models for Spatial and Spatiotemporal
data 409
16.4.1 Simulation Study 411
16.5 P-spline Functional Spatial Regression 413
16.6 Application to Air Pollution Data 415
16.6.1 Spatiotemporal Smoothing 416
16.6.2 Spatial Functional Regression 416
Acknowledgments 421
References 421
Index 424
xiii
List of Contributors
Ana M. Aguilera María del Carmen Aguilera Morillo

University of Granada Universitat Politècnica de València
Department of Statistics and Department of Statistics and
Operational Research Operational Research and Quality
Spain Spain
Mohamed-Salem Ahmed Sophie Dabo-Niang

University of Lille University of Lille
France France
Mara S. Bernardi Maria Durban

Politecnico di Milano Universidad Carlos III
MOX - Department of Mathematics Department of Statistics
Italy Spain
John Ensley
Gregory Bopp
Pennsylvania State University
Department of Statistics
Department of Statistics
USA
USA
Maria Franco-Villoria
Martha Bohorquez Università di Modena e Reggio Emilia
National University of Colombia Department of Economics
Department of Statistics “Marco Biagi”
Colombia Italy
Laurence Broze Zied Gharbi

University of Lille University of Lille
France France
xiv List of Contributors
Ramón Giraldo Dae-Jin Lee

National University of Colombia BCAM–Basque Center for Applied
Department of Statistics Mathematics
Colombia Spain
Alberto Guadagnini Claude Manté

Politecnico di Milano Université du Sud Toulon-Var
Department of Civil and CNRS/INSU, IRD, MIO, Aix-Marseille
Environmental Engineering Université
Italy France
and
Jorge Mateu
The University of Arizona University Jaume I of Castellon
Department of Hydrology and Department of Mathematics
Atmospheric Sciences Spain
USA
Alessandra Menafoglio
Rosaria Ignaccolo Politecnico di Milano
Università degli Studi di Torino MOX - Department of Mathematics
Dipartimento di Economia e Statistica Italy
“Cognetti de Martiis”
Italy Pascal Monestiez
INRAE - Unité BioSP
Antonio Irpino France
University of Campania “Luigi
Vanvitelli” David Nerini
Department of Mathematics and Université du Sud Toulon-Var
Physics CNRS/INSU, IRD, MIO, Aix-Marseille
Italy Université
France
Piotr Kokoszka
Colorado State University Federica Passamonti
Department of Statistics Politecnico di Milano
USA MOX - Department of Mathematics
Italy
Sara Sjöstedt de Luna
Umeå University Davide Pigoli
Department of Mathematics and King’s College London
Mathematical Statistics UK
Sweden
List of Contributors xv
Alessia Pini Johan Strandberg

Università Cattolica del Sacro Cuore Umeå University
Department of Statistical Sciences Department of Statistics
Italy Sweden
Cristian Preda Camille Ternynck

Institute of Statistics and Applied University of Lille
Mathematics of the Romanian France
Academy
Romania Baba Thiam
University of Lille
Matthew Reimherr France
Department of Statistics Vincent Vandewalle
USA University of Lille
France
Elvira Romano
University of Campania “Luigi Simone Vantini
Vanvitelli” Politecnico di Milano
Department of Mathematics and MOX - Department of Mathematics
Physics Italy
Italy
Valeria Vitelli
Laura M. Sangalli University of Oslo
Politecnico di Milano Oslo Center for Biostatistics and
MOX - Department of Mathematics Epidemiology
Italy Department of Biostatistics
Norway
Piercesare Secchi
Politecnico di Milano Anne-Françoise Yao
MOX - Department of Mathematics Université Clermont-Auvergne
Italy France
and
CADS - Center for Analysis Decisions
and Society
Human Technopole
Italy
xvi
Foreword
Functional data analysis (FDA) is a branch of statistics that analyses data providing
information about curves, surfaces, or anything else varying over a continuum.
In its most general form, under an FDA framework each sample element is a func-
tion. The continuum over which these functions are defined is often time, but may
also be spatial location, wavelength, probability, etc. In the 20 years since the first
books and papers on this topic, this field of statistics has received the attention and
encouragement of researchers in statistics and many applied disciplines and has
become an important and dynamic area of modern statistics. Topics that have been
covered include descriptive techniques, statistical inference, multivariate and non-
parametric methods, regression, generalized linear models, time series, and spatial
statistics.
Modern technology has made it possible to obtain large spatial and spatiotem-
poral data sets, and poses the challenge of statistical modeling of such data. The
combination of spatial statistics with FDA has emerged as a key approach. This
book presents new theories and methods to define, describe, characterize, and
model functional data indexed in spatial or spatio-temporal domains. The main
focus is on functional data obtained under a geostatistical framework, where the
domain is fixed and continuous. Specific topics considered include kriging, clus-
tering, regression, and optimal sampling, moving on in the last part of the book to
spatiotemporal data. Some chapters also consider the treatment of functional data
on lattices.
When we wrote our original book on the subject in the 1990s, James Ramsay and
I hoped that we would encourage FDA as a way of thinking, not simply a collection
of techniques. It has therefore been very pleasing to see the development of the
field since then, and the abundance of research activity in the area has confirmed
our hopes. I would urge readers and researchers to raise their sights above any
specific methods, obviously important that they are, to ask how considering data
as functions changes and broadens our statistical horizons. Particularly in the new
era of data science, this concerns both what data can be collected and how they
can be analyzed. I am sure this book will make a valuable contribution in helping
them to do so.
November 2020 Sir Bernard Silverman

University of Oxford
University of Nottingham
Oxford and Nottingham
1
Introduction to Geostatistical Functional Data Analysis

Jorge Mateu 1 and Ramón Giraldo 2
1
Department of Mathematics, University Jaume I of Castellon, Spain
2
Department of Statistics, National University of Colombia, Bogota, Colombia
1.1 Spatial Statistics
Spatial statistics has developed rapidly during the last 30 years. We have seen
an interesting progress both in theoretical developments and in practical stud-
ies. Some early applications were in mining, forestry, and hydrology. It seems to
be honest to remark that the increasing availability of computer power and skill-
ful computer software has stimulated the ability to solve increasingly complex
problems. Clearly, these problems have some common elements: they were all
of a spatial nature. Some theory was available, for example the random function
theory as developed by Yaglom and others in the 1960s. But that was largely insuf-
ficient to find generic solutions for the whole class of problems, and hence, the
applications required a new theory. Thereupon some far-reaching theories have
been developed: image reconstruction, Markov random fields, point process statis-
tics, geostatistics, and random sets, to mention just a few. As a next stage, these
theories were applied successfully to new disciplinary problems leading to mod-
ifications and extensions of mathematical and statistical procedures. We there-
fore notice a general scientific process that has occurred in the field of spatial
statistics: well-defined problems with a common character were suddenly on the
agenda, and data availability and intensive discussion with practical and disci-
plinary researchers resulted in new theoretical developments. Often, it is difficult
to say which was first, and what followed, but we see different theoretical models
developed for different applications.
Spatial statistics has hence emerged as an important new field of science. One
of the peculiarities is its power for visualization. A common cold-water fear of
many statisticians and mathematicians to analyze images, to communicate their
Geostatistical Functional Data Analysis, First Edition. Edited by Jorge Mateu and Ramón Giraldo.
© 2022 John Wiley & Sons Ltd. Published 2022 by John Wiley & Sons Ltd.
2 1 Introduction to Geostatistical Functional Data Analysis
results by maps, and to have to trust information in pictures was overcome. It has
led to interesting theories and better and more objective procedures for dealing
with spatial variation. Following Wittgenstein, we could state that we needed some
geniuses to tackle the obvious. Now, many results of a spatial statistical analysis
could be communicated smoothly toward the nonstatistical audience, like a disci-
plinary scientist, a policy-maker, or an interested student. They, in turn, were able
to judge whether a problem was solved, whether a policy measure was relevant or
was inspired by the beautiful pictures expressing deep thoughts on relevant issues.
The role in policy-making may be once more stressed. It is known that many
policy-makers are inclined to make a decision on the basis of a well developed,
well organized, and well understandable figure. They find it (rightly so!) rather
boring to use long lists of statistical data. But as political decisions affect us all, it
puts another responsibility on the back of statisticians: to make statistically sound
maps. It is often hard to say what that should be, but at the very least, we should be
able to generate pictures, maps, and graphs that rely on good data and that show
important aspects for decision-making.
In this way, spatial statistics has become a refreshing wind in statistics. We do
not need to do well much longer on difficult equations, long lists of data, and tables
with simulated controlled scenarios. But, to be clear on the back of all these nice
pictures a sound science with sometimes difficult and tedious derivations and deep
thoughts are still required to make serious progress.
Spatial statistics recognizes and exploits the spatial locations of data when
designing for, collecting, managing, analyzing, and displaying such data. Spatial
data are typically dependent, for which there are classes of spatial models
available that allow process prediction and parameter estimation. Spatially
arranged measurements and spatial patterns occur in a surprisingly wide variety
of scientific disciplines. The origins of human life link studies of the evolution of
galaxies, the structure of biological cells, and settlement patterns in archaeology.
Ecologists study the interactions among plants and animals. Foresters and agri-
culturalists need to investigate plant competition and account for soil variations
in their experiments. The estimation of rainfall and of ore and petroleum reserves
is of prime economic importance. Rocks, metals, and tissue and blood cells
are all studied at a microscopic level. Geology, soil science, image processing,
epidemiology, crop science, ecology, forestry, astronomy, atmospheric science, or
simply any discipline that works with data collected from different spatial loca-
tions, need to develop models that indicate when there is dependence between
measurements at different locations. Spatiotemporal variability is a relatively new
area within Spatial Statistics, which explains the scarcity of space-time statistical
tools 20 years ago. There has been a growing realization in the last decade that
knowing where data were observed could help enormously in answering the
substantive questions that precipitated their collection. One of the most powerful
tools for spatial data analysis is the map. For example, in military applications,
the battlespace is mapped for command and control. The sensors are both in situ
and remote, and they generate spatially distributed data of many different kinds.
Producing a statistically optimal map, together with measures of map uncertainty,
which is always up to date, is a complicated task. Once these types of statistical
problems are solved, a geographic information system, or GIS, is well suited to
forming the decision-making maps.
Spatial statistics can be considered a natural generalization of signal processing
to higher dimensions. In traditional signal processing, one has a signal dependent
on a scalar variable t, which may belong to a discrete set or which may be contin-
uous. Spatial statistics is concerned with cases in which t is a multidimensional
index of dimension d > 1. In most practical examples d = 2, though much of the
basic theory and methodology is the same whatever the dimension. Although the
models and methods of spatial statistics have not developed as rapidly as those for
one-dimensional signal processing, there have nevertheless been substantial new
developments in recent years. Standard and modern references on spatial statistics
include the books of [1–4] among others.
Following Cressie [5], spatial data can be thought of as resulting from observa-
tions on the stochastic process {Z(s) ∶ s ∈ D}, where D is possibly a random set in
ℝd . If we believe that the roots of statistical science are in data, we can classify spa-
tial areas according to the type of observations encountered. Thus, (i) if D is a fixed
subset of ℝd and Z(s) is a random vector at location s ∈ D, we are dealing with geo-
statistical data; (ii) if D is a fixed (regular or irregular) collection of countably many
points of ℝd and Z(s) is a random vector at location s ∈ D, we are dealing with lat-
tice data; (iii) if D is a point process in ℝd and Z(s) is a random vector at location
s ∈ D, we are dealing with point patterns; (iv) if D is a point process in ℝd and
Z(s) is itself a random set, we are dealing with spatial objects. Geostatistical-type
problems are distinguished most clearly from lattice-and point-pattern-type prob-
lems by the ability of the spatial index s to vary continuously over a subset of ℝd .
A space-time process can be denoted by {Z(s, t) ∶ s ∈ D(t), t ∈ T}, where each of
Z, D, and T is possibly random.
Spatial statistics is one of the major methodologies of environmental statistics.
Its applications include producing spatially smoothed or interpolated representa-
tions of air pollution fields, calculating regional average means or regional average
trends based on data at a finite number of monitoring stations, and performing
regression analyses with spatially correlated errors to assess the agreement
between observed data and the predictions of some numerical model. The notion
of proximity in space is implicitly or explicitly present in the environmental
sciences. Proximity is a relative notion, relative to the spatial scale of the scientific
investigation. When a spatial dimension is present in an environmental study, the
statistician’s job is to create a statistical framework within which one carries out
defensible inferences on processes and parameters of interest. These modeling

and inference strategies are not always easy to do, but are never impossible. If
statistics is to continue to be the broker of variability, it must address difficult
questions such as those found in the environmental sciences, otherwise, it will
become marginalized as a discipline. Problems in the environmental sciences are
inherently spatial (and temporal), observational in nature, and have experimental
units that are highly variable.
In the last decade, spatial statistics has undergone enormous development in
the area of statistical modeling. It started slowly, building from models that were
purely descriptive of spatial dependence. Then, it became apparent that the pro-
cess of interest was usually hidden by measurement error and that the principal
goal should be inference on the hidden process from the noisy data. It has only
been in the last few years that the full potential for hierarchical spatial statistical
modeling has been glimpsed. There is an enormous amount of flexibility in hier-
archical statistical models, such as the opportunity to account for nonlinearities.
Their attractive feature is that at each level of the hierarchy, the model specifica-
tion is simple, yet globally, the model can be quite complex. This approach could
be summarized as a model locally, analyze globally.
Applications of spatial statistics cover many areas. Much of the original impetus
for the subject was driven by geostatistics. It was in this context that the tech-
nique of kriging, optimal least squares interpolation over a random spatial field,
was originally developed. In recent years, the applications of spatial statistics have
increased enormously, with particularly fruitful applications in the environmen-
tal and ecological sciences. A typical problem is the sampling of a pollution field,
such as ozone in the atmosphere or toxic chemicals in rivers and lakes. Another
example is the use of meteorological measurements in studies of global climate
change. In these fields, as in geostatistics, the objective may be to interpolate spa-
tially between measurements, but there are also other objectives which may be
quite different. Spatial statistics has also found applications in such diverse fields
as sociology, for example social networks theory and financial economics.
The usual approach in geostatistics is based on an assumption that the spatial
random field is stationary and isotropic. In the original geophysical applications
which motivated the development of the field, this assumption was often justified
by the fact that with sparse data, there was no reasonable alternative. A further
point is that many geostatistical applications involved only one measurement at
each site (or equivalently, only one replication of the random field) so there was no
way of determining the complete spatial covariance function without some kind
of stationarity assumption. In modern environmental applications, however, there
are very often enough monitoring stations to go beyond such assumptions, and
with multiple observations per site, it is also possible to estimate the covariance
between any pair of sites without assuming stationarity across the field. Another
consideration is that very often, simple topography makes a stationary assumption

implausible. Therefore, there are by now many reasons to go beyond a stationary
model. In spite of this obvious need for nonstationary models; however, there is
not, as yet, a wide variety of approaches to the problem.
Environmental issues have brought atmospheric science to the center of
science and technology, where it now plays a key role in shaping national and
international policy. Weather prediction plays a significant role in the planning of
human affairs. Further, a broader appreciation of the role of weather and climate
impacts on the environment of the planet has now led to nearly universal concern
regarding potential climate change, its causes, impacts, and possible remedying.
A large variety of statistical methods are used routinely in the atmospheric
sciences. For example, techniques of multivariate time series are especially
common. These include multivariate autoregressive, moving average models and
Kalman filtering. Statistical methods for spatial data are also standard. A major
tool in the analysis of space-time data is empirical orthogonal functions (EOF).
Virtually, all atmospheric and oceanographic processes (e.g. wind, temperature,
sea surface temperature, moisture) involve variability over space and time.
One only needs examine the governing partial differential equations for wind
processes, or their selected spatial-temporal averages, to see that mathematical
and statistical descriptions of these dynamical processes depend on complicated
temporal and spatial relationships. Furthermore, observations of geophysical
processes typically include measurement errors and are often temporally and
spatially incomplete, which may obscure the signal of interest.
In studies involving spatial data, it is seldom the case that data for only a single
process are collected. Typically, there is a great expense associated with establish-
ing spatial monitoring networks or other mechanisms of spatial data collection
(e.g. satellites) and so measurements are usually made on two or more variables.
Thus, statistical techniques for multivariate spatial data are critical for effective
modeling of spatial processes.
Lately, there has been a rich and growing literature on space–time model-
ing. Fundamentally, it is clear that in the absence of a temporal component,
second-order geostatistical models can be used to represent spatial variability.
These are descriptive in the sense that, although they model spatial correlation,
there is no causative interpretation associated with them. Thus, for space-time
modeling, the geostatistical paradigm assumes a descriptive structure for both
space and time (i.e. covariance structures are directly specified). For example,
one can extend the geostatistical kriging methodology for spatial processes by
assuming that time is just another spatial dimension. Alternatively, one can treat
time slices of a spatial field as variables and apply a multivariate or cokriging
approach. Although these approaches have been successful in many applications,
there are fundamental differences between space and time, and it is not likely
that realistic covariance structures can be specified that accurately capture the
complicated dynamical processes as found in geophysical applications.
In the absence of a spatial component, there is a large class of time series models
that could be used to represent the temporal variability. These are dynamic in the
sense that they exploit the fact that time flows in only one direction, and so the
state of the process at the current time is related to what happened at previous
times. Thus, one might consider the space–time process as a collection of spa-
tially correlated time series in continuous space, or on a spatial lattice. Although
these approaches include dynamical structures, without a descriptive spatial com-
ponent one lacks the ability to perform spatial prediction at locations without
observations. If both temporal and spatial components are present, it is natural to
combine the temporally dynamic state-space approach and the spatially descrip-
tive approach. These models are referred to as space–time dynamic models.
Spatial interpolation is an essential feature of many GIS. It is a procedure
for estimating values of a variable at unsampled locations. A map with iso-
lines is usually the visual output of such a process and plays a crucial role in
decision-making. Based on Tobler’s law of geography, which stipulates that
observations close together in space are more likely to be similar than those
farther apart, the development of models attempting to represent the way close
observations are related can sometimes be very problematic. The approaches can
be divergent and may therefore lead to very different results. As a consequence,
an understanding of the initial assumptions and methods used is the key to the
spatial interpolation process.
Surprisingly, when spatial interpolation tools are integrated within GIS, they are
often implemented in such a way that users have no real choice in selecting the
best possible methods, and if they do have a choice, required input parameters
are sometimes fixed, without any possible way of modifying them. One reason
for the frequent blind use of spatial interpolation methods, and spatial statistics
in general, probably has its origins in teaching. Despite the large variety of its
applications, the discipline has been confined to those fields where it has seen its
major developments. The progress made in spatial statistics is therefore usually
presented only in journals dedicated to statistics, mining, and petroleum engi-
neering. As a consequence, GIS users who have a different technical background
often do not have an in-depth knowledge of such spatial interpolation techniques.
Furthermore, since the conventional tests used in basic statistics usually gener-
ate some kind of categorical answer, the prerequisite experience and statistical
knowledge necessary for the proper use of spatial interpolation techniques are
often discouraging to this type of users. Nevertheless, during the last few years,
the diversity of the applications of these methods has encouraged the publication
of new books and new case studies and has stimulated a number of conferences
on the subject.
1.2 Spatial Geostatistics

This section has been partially taken and summarized in parts from [6], intending
to provide a brief overview to spatial geostatistics. The reader is referred to [6] for
further and more complete details.
1.2.1 Regionalized Variables

Geostatistics can be defined as the study of regionalized phenomena, that is, phe-
nomena that stretch across space and which have a certain spatial organization or
structure. However, geostatistics is not applied to the regionalized phenomenon as
such, which is a physical reality, but to a mathematical description of that reality,
that is, a numerical function called regionalized variable or regionalization, defined
in a geographical space, which is supposed to correctly represent and measure that
phenomenon.
In order to delve deeper into the concept of regionalized variable, let us imagine
we are interested in a feature of a given phenomenon that spans across space and
that several measurements are taken in a domain D at a given moment in time.
If the measurements are taken on objects or similar, the objects sampled can be
considered a subset of a larger collection of objects, as many more measurements
could have been taken, but were not for many possible reasons. If the observations
were made at certain points in the domain, infinite measurements could be taken.
When s spans across the domain under study, D, the set {z(s), s ∈ D}, is called
a regionalized variable or regionalization, the set {z(si ), i = 1, 2, 3, …} being a
collection of values of the regionalized variable, and each value of that collection
being a regionalized value.
It is true that a deterministic approach can be employed to describe or model
a regionalized phenomenon and obtain an accurate assessment of the values of
the regionalization on the basis of a limited number of observations. However,
this requires in-depth knowledge of the origin of the phenomenon and the phys-
ical or mathematical laws that govern the evolution of the regionalized variable.
Furthermore, many of the regionalized phenomena that are usually studied are
so complex that a deterministic approach can only partially portray them. That is
why the deterministic approach is discarded and the probabilistic approach, which
permits modeling both the knowledge of and also the uncertainty surrounding the
regionalized random phenomenon, is adopted.
1.2.2 Random Functions

From a probabilistic perspective, the regionalized value can be seen as the result
of a random mechanism, resulting in a random variable (r.v.). If the regionalized
values at all the points in the domain D are considered, it can be seen as a reality
of an infinitely large set of r.v.s, one at each point in the domain, which is known
as spatial random function (synonyms: stochastic process, random field).
When s spans across the domain under study, D, we have a family of r.v.s,
{Z(s), s ∈ D}, which constitutes a spatial random field (r.f.).
This methodological decision is one of the cornerstones of geostatistics: the
regionalized variable is interpreted as a realization of a spatial r.f. At this point,
we must state that the regionalized variable is often highly locally irregular
(which makes it impossible to represent using a deterministic mathematical
function) and has a certain spatial organization or structure. The probabilistic
approach, or probabilistic geostatistics, which interprets the regionalized variable
as a realization of a r.f., can take into account all the aspects of regionalization
mentioned above, because, as stated in page 55 of [7]:
i) At each location s, Z(s) is a r.v. (hence, the erratic aspect).
ii) For any given set of points s1 , s2 , … , sk , the r.v.s Z(s1 ), Z(s2 ), … , Z(sk ) are linked
by a network of spatial correlations responsible for the similarity of the values
they take (hence the structured aspect).
Let Z(s) be a r.f. and let us consider the set of points (s1 , … , sk ). Then, the
r.f. Z(s) is characterized by its k-dimensional distribution function. The set of
k-dimensional distribution functions for all values of k and all possible choices of
(s1 , … , sk ) in the domain is called the spatial law of probability.
For a given r.f., Z(s), the k-dimensional distribution function F(z(s1 )), … , (z(sk ))∶
ℝd → [0, 1] is defined as
FZ1 ,…,Zk (z(s1 )), … , (z(sk )) = P[Z(s1 ) ≤ z(s1 ), … , Z(sk ) ≤ z(sk )]. (1.1)
In linear geostatistics, it is enough to know the first two moments of the distribu-
tion of Z(s). What is more, in most practical applications, the available information
does not allow to infer higher-order moments.
The expectation, expected value or first-order moment of a r.f. is defined as a
nonrandom function of s that coincides at each point with the expectation of the
r.v. at that point 𝜇(s) = E(Z(s)), where 𝜇(si ) = E(Z(si )), ∀i ∈ ℕ. It is also called the
drift of the r.f., especially when it varies with location.
The variance of a r.f. is defined as a nonrandom function of s that coincides at
each point with the variance of the r.v. at that point, i.e. V(s) = V(Z(s)), where
V(si ) = V(Z(si )), ∀i ∈ ℕ.
The covariance function of a r.f. is defined as a nonrandom function of si and sj ,
such that for any pair of values (si , sj ) coincides with the covariance between the
r.v. at those two points
( )
C(si , sj ) = C(Z(si ), Z(sj )) = E (Z(si ) − 𝜇(si ))(Z(sj ) − 𝜇(sj )) , ∀si , sj ∈ D.
(1.2)
The variogram of the r.f. is defined as the variance of the first differences of the r.f.
( )
2𝛾(si − sj ) = V (si ) − Z(sj ) , ∀si , sj ∈ D. (1.3)
The function 𝛾(si , sj ) is called semivariogram.

Z(s) is a Gaussian r.f. if for all k and any given set of points s1 , … , sk , the joint dis-
tribution of Z(s1 ), … , Z(sk ) is a multivariate Gaussian distribution. A multivariate
Gaussian distribution is characterized by a mean vector and a variance–covariance
matrix, such that the two first moments of a Gaussian r.f. completely determine
its probability structure. The Gaussianity of the r.f. is a common assumption in
geostatistics.
1.2.3 Stationarity and Intrinsic Hypothesis

regionalized variable in probabilistic terms as a particular realization of a given r.f.
{Z(s), s ∈ D} makes operational sense when it is possible to infer part or all of the
law of probability which defines that r.f. In this sense, stationarity, which indicates
a certain degree of homogeneity in the regionalization across space, is a desirable
quality.
Indeed, it would be impossible to infer the probability law of a r.f. if there was
only one realization of the r.f. In order to make inferences consistently, many real-
izations are necessary. However, in reality there is only one. The solution to this
problem is to adopt the hypothesis of stationarity or spatial homogeneity. The idea
behind the hypothesis of stationarity is to substitute repetitions of the (inaccessi-
ble) realizations of the r.f. with repetitions in space, that is, the values observed at
different locations in the domain under study have the same characteristics and
can be considered as realizations of the same r.f. in mathematical terms. However,
these realizations are not independent, and an additional hypothesis, ergodicity,
is normally assumed; see pages 19–22 of [8] for details. The hypothesis of station-
arity means that the spatial law of probability of the r.f. or part of it, is transla-
tion invariant. That is, the probabilistic properties of a set of observations do not
depend on the specific locations where they have been measured, but only on their
separations.
Therefore, in mathematical and probabilistic terms, the hypothesis of stationar-
ity refers to the regular behavior in space of the moments of the r.f., or the func-
tion itself and, as we will see later, there are different degrees of stationarity. This
hypothesis will allow us to act as if all the variables that make up the r.f. had
the same probability distribution (or the same moments; we can even relax this
assumption) and, as a consequence, to be able to make inferences.
Using the assumed level of spatial homogeneity of the r.f. that (supposedly) gen-
erates the observed realization as a basis, we have the following cases: Stationary
random function in the strict sense, second-order stationary random function, and
intrinsically stationary random function or random function of stationary incre-

ments. Let us briefly introduce these concepts.
The r.f. {Z(s), s ∈ D} is said to be stationary in the strict sense, or strictly sta-
tionary, if the families of r.v.s Z(s1 ), … , Z(sk ) and Z(s1 + h), … , Z(sk + h) have the
same joint distribution function for all k, and for any given spatial points and any
translation vector h ∈ ℝd .
In other words, the joint distribution function of {Z(s1 ), Z(s2 ), … , Z(sk )} is unaf-
fected by the translation of an arbitrary quantity h. As a result, density functions
with dimension lower than k do not depend on location either. Generally speak-
ing, this is a strongly strict condition, which is why this hypothesis is normally
relaxed to the so-called “assumption of second-order stationarity,” which limits
the stationarity hypothesis to the first two moments of the r.f. (recall that in linear
geostatistics, we are only interested in the two first moments of the r.f.).
The r.f. {Z(s), s ∈ D} is said to be second-order stationary, weakly stationary or
stationary in the broad sense, if it has finite second-order moments (that is the
covariance exists) and verifies that
● The expectation exists and is constant, and therefore does not depend on the
location s
E (Z(s)) = 𝜇(s) = 𝜇. (1.4)
● The covariance exists for every pair of r.v.s, Z(s) and Z(s + h), and only depends
on the vector h that joins the locations s and (s + h)
C (Z(s), Z(s + h)) = C(h), ∀s ∈ D and h. (1.5)
As the covariance function C(h) of a second-order Stationary, r.f. is only a func-

tion of h, the variance of the r.f. exists and is finite and constant:
V(Z(s)) = C(𝟎) = 𝜎 2 . (1.6)
In light of Eqs. (1.4) and (1.6), the second-order stationarity hypotheses can
be interpreted as if the regionalized variable takes values that fluctuate around
a constant value (the mean), and the variation of these fluctuations is the same
everywhere in the domain.
In some cases, in order to model the spatial dependence of second-order station-
ary r.f.s, the correlogram, or correlation function, is used instead of the covariogram,
and is defined as
C(h)
Corr(Z(s), Z(s + h)) = = 𝜌(h). (1.7)
C(𝟎)
In case of second-order stationarity, the covariance function and the semi-
variogram are equivalent when it comes to defining the structure of spatial
dependence displayed by the phenomenon, as they verify the following mutual

relationship:
1
𝛾(h) = V (Z(s + h) − Z(s))
2
1
= (V(Z(s + h)) + V(Z(s)) + 2C(Z(s + h), Z(s)))
2
1
= (C(𝟎) + C(𝟎) + 2C(h))
2
= C(𝟎) − C(h).
Notice that if a r.f. is strictly stationary, then it is also stationary in the broad
sense. The converse, however, is generally not true. Obviously, for Gaussian r.f.s,
second-order stationarity is equivalent to strict stationarity.
A r.f. is said to be quasistationary when the corresponding stationary hypothesis
(usually, the hypothesis of second-order stationarity) is valid only for distances
|h| < d, where d is a limit distance. That is, in the second-order quasistationary
case (usually referred as the quasistationary case) 𝜇(s + h) ≈ 𝜇(s) if |h| < d and
C (Z(s + h) − Z(s)) = C(h) if |h| < d.
Second-order stationarity can also be considered a strict assumption on many
occasions, as it implies the existence of the variance in the r.f. A phenomenon
may have infinite variation capacity and be impossible to model using a r.f. with
finite variance. However, there are cases in which the increments or differences
Z(s + h) − Z(s) do have finite variance and, therefore, are second-order stationary.
This type of r.f. is described as being intrinsically stationary.
The r.f. {Z(s), s ∈ D} is said to be intrinsically stationary (or simply intrinsic) if,
for any given translation vector h, the first-order increments Z(s + h) − Z(s) are
second-order stationary, that is,
E (Z(s + h) − Z(s)) = 𝜇(s), (1.8)
where 𝜇(s), the drift, is necessarily linear in h, and
( )
C (Z(s + h) − Z(s)), (Z(s + h + h′ ) − Z(s + h′ )) = C(h, h′ ), (1.9)
which is equivalent to
1
V (Z(s + h) − Z(s)) = 𝛾(h), (1.10)
2
which is only a function of h.
Obviously, in case that the linear drift is zero
E (Z(s + h) − Z(s)) = 0
and
E(Z(s + h) − Z(s))2 = 𝛾(h),
If a r.f. is second-order stationary, then it is also intrinsically stationary. How-

ever, the converse is not necessarily true. Intrinsic r.f.s that are not second-order
stationary are called strictly intrinsic r.f.s. In particular, a Gaussian intrinsic r.f. is
an intrinsic r.f. whose increments follow a multivariate Gaussian distribution.
A r.f. is said to be quasiintrinsic when the intrinsic hypotheses is valid only for
distances |h| < d, where d is a limit distance.
A r.f. {Z(s), s ∈ D} for which the mean and/or the covariance function depends
on the location (are not translation invariant), is said to be a nonstationary r.f.
When a r.f. {Z(s), s ∈ D} has a drift, i.e. its mean is nonconstant and varies with
location, and its first-order increments Z(s + h) − Z(s) are nonstationary, it is said
that the r.f. is a nonintrinsic r.f. (some authors call them intrinsic random functions
of order k > 0).
1.3 Spatiotemporal Geostatistics

This section has been partially taken and summarized in parts from [9], intending
to provide a brief overview to spatiotemporal geostatistics. The reader is referred
to [9] for further and more complete details.
Geostatistical research has typically analyzed r.f.s, in which every spatiotempo-
ral location can be seen as a point on ℝd × ℝ. While from a mathematical point
of view ℝd × ℝ = ℝd+1 , from a physical perspective, it would make no sense to
consider spatial and temporal aspects in the same way, due to the significant dif-
ferences between the two axes of coordinates. Therefore, while the time axis is
ordered intrinsically (as it exists in the past, present, and future), the same does
not occur with the spatial coordinates.
Assume that observations stem from a r.f. given by Z(x, s, t) = 𝜂(x(s, t), s, t, 𝛽) +
𝜖(x, s, t), s ∈ D, t ∈ T, where s denotes a spatial location, t a time point, x some
potentially space and time-dependent regressors, 𝜂 a parametrized trend model,
D ⊂ ℝd (very often d = 2), and T ⊂ ℝ. For ease of notation, we remove the term
in the covariates x, and write Z(s, t), assuming whenever necessary that any trend
coming from a set of covariates has already been removed.
1.3.1 Relevant Spatiotemporal Concepts

A spatiotemporal r.f. Z(s, t) is said to be Gaussian if the random vector
Z = (Z(s1 , t1 ), … , Z(sn , tn ))′ for any set of spatiotemporal locations follows a
multivariate normal distribution. When not stated explicitly, the indexes i and j
will go from 1 to n.
The spatiotemporal r.f. Z(s, t) is said to have a spatially stationary covari-
ance function if, for any two pairs (si , ti ) and (sj , tj ) on ℝd × ℝ, the covariance
C((si , ti ), (sj , tj )) only depends on the distance between the locations (si and sj ) and
the times ti and tj . And the spatiotemporal r.f. Z(s, t) is said to have a temporarily
stationary covariance function if, for any two pairs (si , ti ) and (sj , tj ) on ℝd × ℝ,
the covariance C((si , ti ), (sj , tj )) only depends on the distance between the times
(ti and tj ) and the spatial locations si and sj . If the spatiotemporal r.f. Z(s, t) has a
stationary covariance function in both spatial and temporal terms, then it is said
to have a stationary covariance function. In this case, the covariance function can
be expressed as
C((si , ti ), (sj , tj )) = C(h, u) (1.11)
with h = si − sj and u = ti − tj the distances in space and time, respectively.
A spatiotemporal r.f. Z(s, t) has a separable covariance function if there is a purely
spatial covariance function Cs (si , sj ) and a purely temporal covariance function
Ct (ti , tj ) such that
C((si , ti ), (sj , tj )) = Cs (si , sj )Ct (ti , tj ) (1.12)
for any pair of spatiotemporal locations (si , ti ) and (sj , tj ) ∈ ℝd × ℝ.
A spatiotemporal r.f. Z(s, t) has a fully symmetrical covariance function if
C((si , ti ), (sj , tj )) = Cs (si , tj )Ct (sj , ti ) (1.13)
for any pair of spatiotemporal locations (si , ti ) and (sj , tj ) ∈ ℝ × ℝ. d
Separability is a particular case of complete symmetry and, as such, any test

to verify complete symmetry can be used to reject separability. In the case of
stationary spatiotemporal covariance functions, the condition of full symmetry
reduces to
C(h, u) = C(h, −u) = C(−h, u) = C(−h, −u), ∀(h, u) ∈ ℝd × ℝ. (1.14)
A spatiotemporal r.f. has a compactly supported covariance function if, for any
pair of spatiotemporal locations (si , ti ) and (sj , tj ) ∈ ℝd × ℝ, the covariance func-
tion C((si , ti ), (sj , tj )) tends toward zero when the spatial or temporal distance is
sufficiently large.
If C(si − sj , ti − tj ) depends only on the distance between positions, that is
( )
∥si − sj ∥, ti − tj , the r.f., apart from being stationary, is also isotropic in space
and time. Note that if the covariance function of a stationary r.f. is isotropic in
space and time, then it is fully symmetrical.
The spatiotemporal variogram is defined as the function
2𝛾((si , ti ), (sj , tj )) = V(Z(si , ti ) − Z(sj , tj )), (1.15)
where V is the variance, and half this quantity is called a semivariogram.
In the case of a r.f. with a zero mean,
2𝛾((si , ti ), (sj , tj )) = E[(Z(si , ti ) − Z(sj , tj ))2 ]. (1.16)
Whenever it is possible to define the covariance function and the variogram,

they will be related by means of the following expression:
2𝛾((si , ti ), (sj , tj )) = V(Z(si , ti )) + V(Z(sj , tj )) − 2C((si , ti ), (sj , tj )). (1.17)
If the spatiotemporal r.f. Z(s, t) has an intrinsically stationary variogram in both
space and time, then it is said to have an intrinsically stationary variogram. In this
case, the variogram can be expressed as
2𝛾((si , ti ), (sj , tj )) = 2𝛾(h, u). (1.18)
The marginals 2𝛾(⋅, u) and 2𝛾(h,⋅) are called purely spatial and purely temporal
variograms, respectively.
A r.f. Z(s, t) is strictly stationary if its probability distribution is translation invari-
ant. Second-order stationarity is a less-demanding condition than strict station-
arity. A spatiotemporal r.f. Z(s, t) is second-order stationary in the broad sense or
weakly stationary if it has a constant mean and the covariance function depends
on h and u.
A spatiotemporal r.f. Z(s, t) is said to be intrinsically stationary if it has a constant
mean and an intrinsically stationary variogram. Intrinsic stationarity is less restric-
tive than second-order stationarity. Another widely used function when modeling
implicit spatiotemporal dependence in a stationary r.f. is the correlation function.
Let Z(s, t) be a second-order stationary r.f. with a priori variance 𝜎 2 = C(𝟎, 0) > 0.
The autocorrelation function of this r.f. is defined as
C(h, u)
𝜌(h, u) = . (1.19)
C(𝟎, 0)
If 𝜌(h, u) is a correlation function on ℝd × ℝ, then its marginal functions 𝜌(𝟎, u)
and 𝜌(h, 0) will, respectively, be the spatial correlation function on ℝd and the
temporal correlation function on ℝ.
A function C((si , ti ), (sj , tj )) of real values, defined on ℝd × ℝ is a covariance func-
tion if it is symmetrical, C((si , ti ), (sj , tj )) = C((sj , tj ), (si , ti )) and positive-definite,
that is,
∑ ∑
n n
ai aj C((si , ti ), (sj , tj )) ≥ 0 (1.20)
i=1 j=1
for any n ∈ ℕ, (si , ti ) ∈ ℝd × ℝ, and ai ∈ ℝ, i = 1, … , n. The condition (1.20) is suf-

ficient if the covariance function can take complex values. Similarly, one necessary
and sufficient condition for a nonnegative function of real values 𝛾((si , ti ), (sj , tj ))
defined on ℝd × ℝ to be a semivariogram is that it is a symmetrical function and
conditionally negative-definite, that is,
∑ ∑
n n
ai aj 𝛾((si , ti ), (sj , tj )) ≤ 0 (1.21)
i=1 j=1
∑n
with i=1 ai = 0.
Schoenberg [10] proved the following theorem characterizing the spatiotem-

poral semivariogram. Let 𝛾((si , ti ), (sj , tj )) be a function defined on ℝd × ℝ, with
𝛾((s, t), (s, t)) = 0, ∀(s, t) ∈ ℝd × ℝ. Then the following statements are equivalent:
● 𝛾((si , ti ), (sj , tj )) is a semivariogram on ℝd × ℝ.

( )
● exp −𝜃𝛾((si , ti ), (sj , tj )) is a covariance function on ℝd × ℝ, for any 𝜃 > 0.
● C((si , ti ), (sj , tj )) = 𝛾((si , ti ), (𝟎, 0)) + 𝛾((sj , tj ), (𝟎, 0)) − 𝛾((si , ti ), (sj , tj )) is a covari-
ance function on ℝd × ℝ.
In case of stationarity, the above results reduce to functions depending on spatial

and temporal lags. Another seminal result that characterizes covariance functions
is that given in [11]. A function C(h, u) defined on ℝd × ℝ is a stationary covariance
function if, and only if, it has the following form
′
C(h, u) = ei(𝛚 h+𝜏u) dF(𝛚, 𝜏), (h, u) ∈ ℝd × ℝ, (1.22)
∫∫
where the function F is a nonnegative distribution function with a finite mean
defined on ℝd × ℝ, which is known as a spectral distribution function. Therefore,
the class of stationary spatiotemporal covariance functions on ℝd × ℝ is identical
to the class of Fourier transforms of nonnegative distribution functions with finite
means on that domain. If the function C can also be integrated, then the spec-
tral distribution function F is absolutely continuous and the representation (1.22)
simplifies to
′
C(h, u) = ei(𝛚 h+𝜏u) f (𝛚, 𝜏)d𝛚 d𝜏, (h, u) ∈ ℝd × ℝ (1.23)
∫∫
where f is a nonnegative, continuous, and integrable function that is known as a
spectral density function. The covariance function C and the spectral density func-
tion f then form a pair of Fourier transforms, and
′
f (𝛚, 𝜏) = (2𝜋)−d−1 e−i(𝛚 h+𝜏u) C(h, u)dh du. (1.24)
∫∫
The decomposition (1.22) can be specialized for fully symmetrical covariance
functions. Let C(⋅, ⋅) be a continuous function defined on ℝd × ℝ, then C(⋅, ⋅) is
a fully symmetrical stationary covariance function if, and only if, the following
decomposition is possible
C(h, u) = cos(𝛚′ h) cos(𝜏u)dF(𝛚, 𝜏), (h, u) ∈ ℝd × ℝ, (1.25)

∫∫
where F is the nonnegative and symmetrical spectral distribution function defined
on ℝd × ℝ.
Cressie and Huang [12] provide a theorem for characterizing the class of sta-
tionary spatiotemporal covariance functions under the additional hypothesis of
integrability. Let C(⋅, ⋅) be a continuous, bounded, symmetrical, and integrable
function defined on ℝd × ℝ, then C(⋅, ⋅) is a stationary covariance function if, and

only if, in view of u ∈ ℝ,
′
C𝛚 (u) = e−i𝛚 h C(h, u)dh, (1.26)
∫
is a covariance function for every 𝛚 ∈ ℝd except, at the most, in a set with a null
Lebesgue mean. Gneiting [13] generalizes this result for C defined on ℝd × ℝl ,
from which the previous statement is a particular case for l = 1.
Both the covariance function and the spectral density function are important
tools for characterizing random stationary spatiotemporal fields. Mathematically
speaking, both functions are closely related as a pair of Fourier transforms.
Furthermore, the spectral density function is particularly useful in situations
where there is no explicit expression of the covariance function. Stein [14] shows
the benefit of using smooth covariance functions far from the origin, which can
be tested by verifying whether their spectral densities have derivatives of certain
orders.
1.3.2 Spatiotemporal Kriging

Kriging is aimed at predicting an unknown point value Z(s0 , t0 ) at a point (s0 , t0 )
that does not belong to the sample. To do so, all the information available about
the regionalized variable is used, either at the points in the entire domain or in a
subset of the domain called neighborhood.
Assume that the value of the r.f. has been observed on a set of n spatiotemporal
locations {Z(s1 , t1 ), … , Z(sn , tn )}. If we want to predict the value of the r.f. on a new
spatiotemporal location (s0 , t0 ), we use the linear predictor
∑
n
Z ∗ (s0 , t0 ) = 𝜆i Z(si , ti ) (1.27)
i=1
constructed from the random variables Z(si , ti ). As in the spatial case, spatiotempo-
ral kriging equations will depend on the degree of stationarity attributed to the r.f.
that supposedly generates the observed realization. The most widely used kriging
techniques in the spatiotemporal case are simple spatiotemporal kriging, ordi-
nary spatiotemporal kriging, and universal spatiotemporal kriging. In the case of
simple spatiotemporal kriging, we assume that Z(s, t) is a second-order stationary
spatiotemporal r.f., with a constant and known mean 𝜇(s, t), constant and known
variance C(𝟎, 0), and a known covariance function C(h, u). The kriging equations
(n equations with n unknown elements) are of the form
∑
n
( ) ( )
𝜆j C si − sj , ti − tj = C si − s0 , ti − t0 , ∀i = 1, … , n (1.28)
j=1
from which we obtain the values 𝜆i that minimize the prediction error variance,
which is given by
[ ] ∑
n
V Z ∗ (s0 , t0 ) − Z(s0 , t0 ) = C(𝟎, 0) − 𝜆i C(si − s0 , ti − t0 ). (1.29)
i=1
In the case of ordinary spatiotemporal kriging, the constant mean 𝜇(s, t) is not
known, and the covariance function C(h, u) is known, under second-order sta-
tionarity. In the case of an intrinsic r.f., the variance is unbounded. In these two
cases, simple kriging cannot be performed as the mean cannot be subtracted. We
must therefore impose a condition of unbiasedness. In these situations, ordinary
spatiotemporal kriging equations can be expressed, in the first case, in terms of
the covariance function, and in the second case, in terms of the semivariogram, as
there is no covariance at the origin.
In the universal kriging approach, assume Z(s, t) is a r.f. with drift, and so the
mean of the r.f. is not constant, but depends on the pairs (s, t). In this situation,
the so-called “condition of unbiasedness” is affected substantially. In this case, the
r.f. can be disaggregated into two components: one deterministic 𝜇(s, t) and the
other stochastic e(s, t) which can be treated as an intrinsically stationary r.f. with
zero expectation, E [e(s, t)] = 0
Z(s, t) = 𝜇(s, t) + e(s, t). (1.30)
We can assume that the mean, even unknown, can be expressed locally by
∑
p
𝜇(s, t) = ah fh (s, t), (1.31)
h=1
{ }
where fh (s, t), h = 1, … , p are p known functions, ah constant coefficients, and
p the number of terms used in the approximation. It must be taken into account
that this expression is only valid locally. In this case, the equations that yield the
prediction of the weights are obtained from the prediction error conditions of zero
expectation and minimum variance.
1.3.3 Spatiotemporal Covariance Models

One key stage in the spatiotemporal prediction procedure is choosing the covari-
ance function (covariogram or semivariogram) that models the structure of the
spatiotemporal dependence of the data. However, while the semivariogram is nor-
mally chosen for this purpose in the spatial case, in a spatiotemporal framework,
the covariance function is the most commonly chosen tool. By referring to a valid
covariographic spatiotemporal model, we are implicitly stating that the covariance
function must be positive-definite. The purely spatial and temporal covariance
models have been widely studied, and there is a long list of those which can be
used to model spatial or spatiotemporal dependence that guarantee the (spatial

or temporal) covariance function is positive-definite. However, this is not the case
in the spatiotemporal scenario, in which constructing valid spatiotemporal covari-
ance models is one of the main research activities. In addition, while it is difficult to
demonstrate that a spatial or temporal function is positive-definite, it is even more
so when seeking to determine valid spatiotemporal covariance models. For this
reason, many authors began to study how to combine valid spatial and temporal
models to obtain (valid) spatiotemporal covariance models.
By way of introduction, the first approximations to modeling spatiotemporal
dependence using covariance functions were nothing more than generalizations
of the stationary models used in the spatial scenario. In this sense, early studies
often modeled the spatiotemporal covariance using metric models by defining a
metric in space and time that allowed researchers to directly use isotropic mod-
els that are valid in the spatial case. Such metric models were characterized by
being nonseparable, isotropic, and stationary. The next step in this initial stage
consisted of configuring spatiotemporal covariance functions by means of the sum
or product of a spatial covariance and a temporal covariance, both of which were
stationary, giving rise to separable, isotropic and stationary models. Later, realiz-
ing the limitations of the two procedures detailed above in terms of capturing the
spatiotemporal dependence that really exists in the large majority of the phenom-
ena studied, interest shifted toward including the interaction of space and time,
in covariance models, giving rise to the so-called “nonseparable models” (while
remaining isotropic and stationary). Development continued with the search for
nonseparable spatiotemporal, spatially anisotropic, and/or temporally asymmet-
rical models, together with general nonstationary models. There is a long list of
papers dealing with these topics in the literature, and here we refer to [6] and all
references therein.
The empirical determination of the covariance function or the variogram of
a spatiotemporal process can be generalized naturally using the procedures for
merely spatial processes. Let Z(⋅, ⋅) be an intrinsically stationary process observed
on a set of n spatiotemporal pairs {(s1 , t1 ), … , (sn , tn )}. Two direct and popular
alternatives to obtain an estimation of the variogram 2𝛾(⋅, ⋅) (and its covariance
function C(⋅, ⋅), if the process is also second-order stationary) are the classical esti-
mator based on the method-of-moments (MoM), and the robust estimator pro-
posed by Cressie and Hawkins [15].
1.4 Functional Data Analysis in Brief

In an increasing number of applications in many disciplines of science, the data
collected corresponds to curves or surfaces. Such data can be generated by repeated
measurements in space–time or by automatic recordings of a quantity of interest.

Functional data analysis (FDA) [16] has been used since the early 1990s to model
this type of information. FDA is a general way of thinking, where the basic unit of
information is an entire function rather than a set of values.
A random variable X(t), t ∈ T ⊂ ℝ is called a functional variable if it takes val-
ues in an infinite-dimensional space (or functional space). A functional random
sample X1 (t), … , Xn (t) corresponds to the collection of n functional variables inde-
pendent and identically distributed (iid) as X(t). An observation of X(t), denoted
as x(t), is called a functional observation. For example, in a medical study xi (t), i =
1, … , n, could represent the electrocardiogram of the ith patient of the sample. In
this case, it is reasonable to assume the independence assumption because it is
natural to think that the responses of the patients are not related. FDA tools allow
to estimate models based on a set of random variables taking values in a space
of functions (functional variables), i.e. it concerns with the statistical analysis of
multiple realizations of one (univariate FDA) or several (multivariate FDA) func-
tional variables. If there is no observational noise, a functional observation xi (t),
t ∈ T, is usually represented as a finite set of pairs (tj , xi (tj )), tj ∈ T, j = 1, … , M.
The set of points {tj }Mj=1
can be considered the same for all the functions in a func-
tional data set, and usually, they form a fine evenly spaced grid in T. Nowadays, the
number M in real applications is usually in the order of several hundred or thou-
sands. Interpolation methods (if there is no noise) or nonparametric smoothing
methods (in the opposite case) are commonly used to represent the data (tj , xi (tj )),
j = 1, … , M, as a real function xi (t). In this sense, we can say that FDA inher-
its methodology from nonparametric estimation. Note that, actually, the curves
are not observed, instead only points of the curves are observed. However, when
the number of points in a curve is dense for simplicity we talk about “observed
curves.”
Since the pioneering work by Deville [17] on harmonic analysis, there has been
a lot of interest in developing statistical models for functional data. Examples of
such methods include exploratory and descriptive data analysis [18], linear models
[19], generalized linear models [20, 21], quantile regression [22], analysis of vari-
ance [23], nonparametric methods [24], longitudinal data [25], additive models
[26] or multivariable techniques [27] such as principal components [28], canonical
correlation [29], discriminant analysis [30] or cluster analysis [31]. An overview of
inference for functional data is shown in [32]. Some new developments in this field
are given in [33] and [34]. A relatively recent problem in FDA is the modeling of
univariate and multivariate misaligned functional data. This one arises when the
functional samples have systematic differences in shape. Some references in this
topic are [35, 36].
Modern technology for acquiring and storing information in real-time often
allows getting data that can be considered as functions. It is also possible to obtain
a finite and therefore incomplete amount of information regarding a function.

For example, when collecting daily temperature data at weather stations. In this
case, it makes sense fitting curves (or surfaces) to obtain functional observations.
Generally, this stage is accomplished using smoothing and nonparametric
methods [37]. This is the first step in FDA. The purpose is to convert discrete
data into a smoothly varying function. In applied FDA, basis functions to obtain
curves from the discrete records (tj , xi (tj )) are generally used [38]. Basis functions
procedures approximate a function by using a fixed truncated basis expansion
∑K
x(t) = l=1 cl Bl (t) = cT B(t) in terms of K known basis functions. Once the
representation by basis functions is adopted, three types of inquiries need to be
answered for computational issues. Which basis functions are appropriate, how
many basis functions are used to fit the data, and how the coefficients of the
vector c are estimated. Generally, Fourier (for periodic data) and B-splines (for
nonperiodic data) basis functions are applied to this purpose [16]. However, other
basis or nonparametric smoothing methods can be used [24]. The number of
basis is estimated by cross-validation. A roughness penalty can be included in the
minimization problem.
Usually, the approaches for modeling functional data are focused on the
assumption that the functions are iid, i.e. it is considered that all the functional
observations correspond to realizations of the same stochastic process. However,
in many fields of science, it is required to model correlated functional data
(temporally or spatially correlated). In these cases, the traditional approaches
based on the iid assumption may not be appropriate. An example of temporally
correlated functional data is that of daily curves of financial transaction data (time
series of functional data). The functions (curves of financial transactions) form
a time series {Xk }; k ∈ ℤ where each Xk is a (random) function Xk (t), t ∈ [a, b]
(a collection of curves temporally indexed). On the other hand, in the spatial
context suppose that penetration resistances (MPa) at different depths (meters)
are recorded in many sites (points) of an experimental farm. The set of curves
Xs(d) , s ∈ ℤ and d ∈ [a, b] define a realization of a spatial stochastic process, where
the response is functional (a collection of curves spatially indexed). Methods for
analyzing correlated data have been adapted to the context of functional data.
Some works in time series analysis and spatial statistics for functional data are
proposed by Zhao et al. [39] and Delicado et al. [40]. A large number of methods
of spatial statistics have been adapted to the functional realm, and this book is a
good and modern example of this. Indeed, when combining spatial (geostatistical)
methods with functional data we enter the field of geostatistical functional data,
which is the core of this book.
FDA has become a rapidly developing discipline in the statistical field given its
wide range of possible applications. Agronomy [41], biology [42], biomedicine
[33, 43], criminology [44], economy [45], medicine [46], meteorology [47],
oceanography [48], psychology [49], and veterinary [50], among others, are
areas where this relatively recent and novel field of statistics is useful. Ullah
and Finch [51] presented a nice compilation of case studies analyzed from a
FDA perspective. To apply FDA to a real dataset, there is a need for appropriate
software with up-to-date methodological implementation and easy addition of
new theoretical developments [52]. R R Core Team [53] and python libraries [52]
can be used for this purpose.
In the geostatistical setting, kriging and cokriging methods have been also
extended to deal with functional data, and methods such as ordinary kriging for
functional data (OKFD), continuous time-varying kriging for functional data
(CTVKFD), and functional kriging total model (FKTM) are now easily available.
The simplest predictor abovementioned is OKFD, where each curve is weighted
by a scalar parameter. The second option (CTVKFD) is founded in the theory
of the functional linear concurrent (pointwise) regression model. In this case,
the parameters are also curves. Finally, the predictor (FKTM) considers double
indexed parameters. Here, in order to carry out the prediction at a particular
time, each observed curve is weighted by a functional parameter. This approach
follows the philosophy of the functional linear total model. Recently, in a similar
framework, these methods have been adapted for Hilbert Spaces.
All the predictors before referenced are based on the assumption that the mean
function is homogeneous into the region of interest. However, in practice, we often
found realizations of nonstationary functional processes (because there exists an
explicit spatial trend). To give solution to the problem of spatial prediction of func-
tional data in the absence of stationarity some alternatives have been proposed. All
of these have arisen as extensions to the functional framework of some classical
kriging methods for nonstationary data.
Geostatistical FDA has been a field of constant growing in the last years. New
modeling requirements in this area are opening many new research avenues. The
definition of predictors considering realizations of multivariate (possibly nonsta-
tionary) functional random fields or the spatial prediction of data belonging to
Riemannian manifolds are only a few new developments which indicate that there
is still a long way ahead in this area of statistics. This book shows a state-of-the-art
with recent contributions in this new environment of the spatial and functional
modeling.
New theories emerge every day that allow adapting and extending traditional
statistical methods to the treatment of functional data. However, there is still a
long way ahead, given the wide range of theoretical and applied possibilities not yet
explored. There are undoubtedly many challenges for the statistical community in
this area.
References
1 Diggle, P. (2013). Statistical Analysis of Spatial and Spatio-Temporal Point

Patterns. CRC Press.
2 Cressie, N. and Wikle, C. (2019). Statistics for Spatio-Temporal Data. Wiley.
3 Diggle, P. and Giorgi, E. (2019). Model-Based Geostatistics for Global Public
Health: Methods and Applications. CRC Press.
4 Wikle, C., Zammit-Mangion, A., and Cressie, N. (2019). Spatio-Temporal Statis-
tics with R. Chapman & Hall/CRC.
5 Cressie, N. (1993). Statistics for Spatial Data. Wiley.
6 Fernandez-Aviles, G., Montero, J., and Mateu, J. (2015). Spatial and
Spatio-Temporal Geostatistical Modeling and Kriging. Wiley.
7 Emery, X. (2000). Geoestadística Lineal. Departamento de Ingeniería de Minas.
Facultad de Ciencias Fisicas and Matematicas. Universidad de Chile.
8 Chiles, J. and Delfiner, P. (1999). Geostatistics: Modeling Spatial Uncertainty.
Wiley.
9 Müller, W. and Mateu, J. (2012). Spatio-Temporal Design. Advances in Efficient
Data Acquisition. Wiley.
10 Schoenberg, I. (1938). Metric spaces and completely monotone functions.
Annals of Mathematics 39: 811–841.
11 Bochner, S. (1933). Monotone funktionen, stiltjes integrale und harmonische
analyse. Mathematische Annalen 108: 378–410.
12 Cressie, N. and Huang, H.C. (1999). Classes of nonseparable, spatiotemporal
stationary covariance functions. Journal of the American Statistical Association
94: 1330–1340.
13 Gneiting, T. (2002). Stationary covariance functions for space-time data. Jour-
nal of the American Statistical Association 97: 590–600.
14 Stein, M. (2005). Space-time covariance functions. Journal of the American Sta-
tistical Association 100: 310–321.
15 Cressie, N. and Hawkins, D. (1980). Robust estimation of the variogram. Jour-
nal of the International Association for Mathematical Geology 12 (2): 115–125.
16 Ramsay, J. and Silverman, B. (2005). Functional Data Analysis. New York:
Springer-Verlag.
17 Deville, J. (1974). Méthodes statistiques et numériques de l’analyse har-
monique. Annales de l’INSEE 15 3–101.
18 Ramsay, J. (2000). Functional components of variation in handwriting. Journal
of the American Statistical Association 95: 9–15.
19 Cardot, H., Ferraty, F., and Sarda, P. (1999). Functional linear model. Statistics
and Probability Letters 45: 11–22.
20 Müller, H. and Stadtmüller, U. (2005). Generalized functional linear models.
Annals of Statistics 33 (2): 774–805.
References 23
21 Febrero-Bande, M. and González-Manteiga, W. (2013). Generalized additive

models for functional data. Test 22 (2): 278–292.
22 Kato, K. (2012). Estimation in functional linear quantile regression. The Annals
of Statistics 40 (6): 3108–3136.
23 Zhang, J. (2014). Analysis of Variance for Functional Data. Boca Ratón, FL:
CRC Press.
24 Ferraty, F. and Vieu, P. (2006). Nonparametric Functional Data Analysis: The-
ory and Practice. Springer Science & Business Media.
25 Park, S. and Staicu, A. (2015). Longitudinal functional data analysis. Stat 4 (1):
212–226.
26 Greven, S. and Scheipl, F. (2017). A general framework for functional regres-
sion modelling. Statistical Modelling 17 (1–2): 1–35.
27 Górecki, T., Krzyśko, M., Waszak, L., and Wołyński, W. (2018). Selected statisti-
cal methods of data analysis for multivariate functional data. Statistical Papers
59 (1): 153–182.
28 Berrendero, J., Justel, A., and Svarc, M. (2011). Principal components for mul-
tivariate functional data. Computational Statistics and Data Analysis 55 (9):
2619–2634.
29 Shin, H. and Lee, S. (2015). Canonical correlation analysis for irregularly and
sparsely observed functional data. Journal of Multivariate Analysis 134: 1–18.
30 Gardner-Lubbe, S. (2020). Linear discriminant analysis for multiple functional
data analysis. Journal of Applied Statistics 48, 1917–1933.
31 Clarkson, D., Fraley, C., Gu, C., and Ramsey, J. (2005). Functional cluster anal-
ysis. S+ Functional Data Analysis: Users Manual for Windows, pp. 155–164.
32 Horváth, L. and Kokoszka, P. (2012). Inference for Functional Data with Appli-
cations, vol. 200. Springer Science & Business Media.
33 Margaritella, N., Inácio, V., and King, R. (2021). Parameter clustering in
Bayesian functional principal component analysis of neuroscientific data.
Statistics in Medicine 40 (1): 167–184.
34 Yu, J., Park, J., Choi, T. et al. (2021). Nonparametric Bayesian functional
meta-regression: applications in environmental epidemiology. Journal of Agri-
cultural, Biological and Environmental Statistics 26 (1): 45–70.
35 Olsen, N., Markussen, B., and Raket, L. (2016). Simultaneous inference for
misaligned multivariate functional data. arXiv preprint arXiv:1606.03295.
36 Zeng, P., Shi, J., and Kim, W.S. (2019). Simultaneous registration and clustering
for multidimensional functional data. Journal of Computational and Graphical
Statistics 28 (4): 943–953.
37 Ferraty, F., Mas, A., and Vieu, P. (2007). Nonparametric regression on func-
tional data: inference and practical aspects. Australian & New Zealand Journal
of Statistics 49 (3): 267–286.
38 Wang, J., Chiou, J., and Müller, H. (2016). Functional data analysis. Annual
Review of Statistics and Its Application 3: 257–295.
39 Zhao, X., Marron, J., and Wells, M. (2004). The functional data analysis view
of longitudinal data. Statistica Sinica 14 789–808.
40 Delicado, P., Giraldo, R., Comas, C., and Mateu, J. (2010). Statistics for spatial
functional data: some recent contributions. Environmetrics: The Official Journal
of the International Environmetrics Society 21(3–4): 224–239.
41 Giraldo, R., Caballero, W., and Camacho-Tamayo, J. (2018). Mantel test for spa-
tial functional data. Advances in Statistical Analysis 102 (1): 21–39.
42 Cremona, M., Xu, H., Makova, K. et al. (2019). Functional data analysis for
computational biology. Bioinformatics 35 (17): 3211–3213.
43 Lopez-Pintado, S. and Qian, K. (2021). A depth-based global envelope test
for comparing two groups of functions with applications to biomedical data.
Statistics in Medicine. https://doi.org/10.1002/sim.8861.
44 Park, K., Suk, H., Hwang, H., and Lee, J.H. (2013). A functional analysis of
deception detection of a mock crime using infrared thermal imaging and the
concealed information test. Frontiers in Human Neuroscience 7: 70.
45 Frois Caldeira, J., Gupta, R., Suleman, M., and Torrent, H. (2020). Forecasting
the term structure of interest rates of the BRICS: evidence from a nonparamet-
ric functional data analysis. In: Emerging Markets Finance and Trade, 1–18.
https://doi.org/10.1080/1540496X.2020.1808458.
46 Strzalkowska-Kominiak, E. and Romo, J. (2021). Censored functional data for
incomplete follow-up studies. Statistics in Medicine. https://doi.org/10.1002/sim
.8930/
47 Suhaila, J., Jemain, A., Hamdan, M., and Zin, W. (2011). Comparing rainfall
patterns between regions in Peninsular Malaysia via a functional data analysis
technique. Journal of Hydrology 411 (3–4): 197–206.
48 Assunção, R., Silva, A., Roy, A. et al. (2020). 3D characterisation of the
thermohaline structure in the southwestern tropical Atlantic derived from
functional data analysis of in situ profiles. Progress in Oceanography 187:
102399.
49 Levitin, D., Nuzzo, R., Vines, B., and Ramsay, J. (2007). Introduction to func-
tional data analysis. Canadian Psychology 48 (3): 135.
50 Gruen, M., Alfaro-Córdoba, M., Thomson, A. et al. (2017). The use of func-
tional data analysis to evaluate activity in a spontaneous model of degenerative
joint disease associated pain in cats. PLoS ONE 12 (1): e0169576.
51 Ullah, S. and Finch, C. (2013). Applications of functional data analysis: a
systematic review. BMC Medical Research Methodology 13 (1): 1–12.
References 25
52 Golovkine, S. (2021). FDApy: a Python package for functional data. arXiv

preprint arXiv:2101.11003.
53 R Core Team (2020). R: A Language and Environment for Statistical Comput-
ing. Vienna, Austria: R Foundation for Statistical Computing. https://www
.R-project.org.
27
Part I
Mathematical and Statistical Foundations

Another random document with
no related content on Scribd:
Venetians should not deliver him up to Spain, deceive him into her power
by the means described.
His affected renunciation of places and profits, together with a shew of
hot persecution from the Spanish and papal court, warranted the friends of
Sebastian in their fatal dependance on his superior talents: he obtained his
victim; and secure of the Duke of Tuscany’s concurrence, apprized him of
the day and hour at which they should enter the gates of Florence.
Juan De Castro, and Don Christopher, who were already in the city, no
sooner heard of their King’s second detention, than aware of the danger
which menaced themselves, and conscious that by the captivity or death of
his friends, Sebastian’s situation would only be rendered more hopeless,
they fled hastily, severally betaking themselves to France and England with
entreaties for effectual aid from both those powers.
Meanwhile Sebastian was reconveyed, with the faithful partner of his
afflictions, from the Florentine prison, to the nearest sea-port, where being
embarked in a Spanish ship of war, and closely kept from the sight of the
crew, they set sail for Naples.
Sebastian rarely condescended to question the only person who was
allowed to attend him, for the man was cold and savage, and seemed
prepared to reply with insults; but on launching again upon that ocean
which he had so often traversed under such variety of fortune, he one day
broke silence, and asked whither they were going to take him. “To the
prison Del Ovo, for life,”—was all the answer of his attendant, as he shut
and bolted the cabin door.
Sebastian and Kara Aziek turned their eyes on each other: they needed
not speech to understand what was passing in each others hearts: their
daughter’s fate alone occupied every feeling.
“Ah, if I could be assured that her innocent life would be spared, her
days pass in peace,” exclaimed Kara Aziek, “my soul would find rest: I
could bound my little remnant of happiness with the walls of my
Sebastian’s prison, or I could die with thee, my husband—die gladly.” She
bent her face on his neck to hide her gushing tears, as she thought of their
perishing together.
Sebastian regarded her tenderly: “I do believe it, my Aziek! cherish this
angel resignation; and since it seems Heaven’s will, that the sins and the
errors of thy husband should descend upon thy guiltless head, O let me
hope, that with so grievous an addition to my burthen as that conviction,
Heaven will be satisfied, and spare me the pang of having caused my
child’s wretchedness.”
To combat this painful and incessant throe of self-condemnation, Kara
Aziek now roused up her fortitude with her love, and while she exhorted
him to remember that human sufferings are much more frequently promised
to the favorites of God, as trials and perfecters, than as penalties and
punishments, her own spirit was elevated and comforted, and she suddenly
appeared endowed with supernatural strength.
Resigned to bear, because humble and confiding, no voice of
lamentation was heard from the chamber of Sebastian and Kara Aziek.
Their dignified stillness, with their gentle and unresisting looks, sometimes
moved even the rough fellow who supplied them with food to murmur as he
left them, “I shall be sorry to hear that they come to harm.”
Sebastian could take no other advantage of this compassion, than that of
winning from his attendant the name of the Neapolitan Viceroy. He learnt
with pleasure that it was the Count of Lemos, a very old and worthy
Spaniard, who had been nobly entertained at the Portuguese court by Don
Sebastian, and had more than once bravely hazarded advice to him on
important subjects, when his own courtiers shrunk from the delicate task.
From a nobleman of this character, both Sebastian and Kara Aziek now
ventured to hope for at least an amelioration of their destiny; and with
something like satisfaction beheld their vessel cast anchor in the bay of
Naples. They were speedily conveyed to the castle Del Ovo, a dark and
fearful fortress, now become a prison for criminals. At sight of the narrow
dungeon, without any other furniture than straw, Kara Aziek’s looks
betrayed the sudden horror with which she was seized, “Is it here we are to
linger out our lives?” she exclaimed, sorrowfully.
“I have no instructions to confine you,” observed the man who had
conducted them, “indeed I never heard of any other prisoner than this
gentleman, so you must be content to abide somewhere else.”
He attempted to take her hand to lead her out, but Kara Aziek sprang
back, and Sebastian advanced to deprecate the heaviest of their misfortunes.
The man urged his orders to confine the pretended Calabrian in a solitary
dungeon; Kara Aziek still resisted, she clung to her husband, wildly
exclaiming:
“Kill me—kill me—tear this poor frame to atoms—still will I remain
here.—Surely no force can take me away, if I am resolved to die beside
him.”
Wrung to torture by her frenzy, Sebastian earnestly sued for permission
to detain her. The man’s inclinations were in favour of compliance, but his
life might have been risked by yielding, and promising to urge their suit in
the morning to the Viceroy’s secretary, he reluctantly repeated his orders.
The arguments and soothings of Sebastian, rather than the explanation
and peremptory behests of the gaoler, allayed the ravings of Kara Aziek;
suddenly she grew calm, started from the ground, and as if alarmed lest her
obstinacy might endanger her husband’s safety, she cried out, “Now, now I
am ready to go!”—a convulsive embrace was exchanged between her and
Sebastian, and the next moment the door of the dungeon closed and divided
them.
“No further!” said she, in a low hurried voice, as the man would have led
her from the spot, “Here is my bed this night—every night—here will I live
till he is restored to me again—force me not from this sad lodging, if you
have love or pity in your heart—I cannot get back to him—I may but hear
his steps and his sighs, and know that he is near me.—Alas! is that too
much of consolation?”
The bitter tears which flowed down her cheeks, and the sorrowful
wringing of her hands, presented so moving a picture, that the Neapolitan
said kindly, “Well, stay here then, I will surely get you admitted in the
morning—what shall I bring you to sleep on?”
“O no sleep—no sleep”—she replied, with joyful wildness, “I will wake
to bless you and to pray for him.” She lightly seated herself on the stone
floor while speaking, and leaning her head against the door of Sebastian’s
cell, remained drinking in at her ear each breath he drew.
Frequently did she long to speak and tell him she was near; but then
conscious that the idea of her being alone and unprotected in an open
passage, exposed to the insults of the wandering guards, and doomed to rest
only on a damp pavement, would overbalance the satisfaction of hearing
her voice, she checked the wish, and relapsed into stillness.
Morning was far advanced when Stephano appeared; he had been to the
secretary and had returned successful. At this intimation, which Kara Aziek
demanded even while he was afar off, she uttered a cry of transport; it was
answered by the voice of Sebastian from within, “Kind heaven! my Aziek,
art thou here again so soon?”
“I have been here the whole night; I would not leave thy door.” While
Aziek was speaking, Stephano unlocked the dungeon, and she flew into the
melancholy, grateful embrace of her husband.
It seemed as if Providence had allotted them this temporary privation
only to make them sensible, that while undivided, they had no right to
abandon themselves to despair. Kara Aziek with overflowing thankfulness
acknowledged this truth, and promised henceforth to grieve no more.—
Stephano passed his hands across his eyes, and replied to some anxious
inquiries of Sebastian.
As it was the most earnest wish of the King to be seen by the Count of
Lemos, he learnt with regret that Lemos was then lying ill of a dangerous
disorder, which devolved his duties upon Sossa, the next nobleman to him
in rank and civil honours. This information was indeed unwelcome;
however, Stephano promised to inquire regularly after the Viceroy’s health,
and to discover whenever his Excellency was in a state to hear of business.
“I am heartily sorry,” he added, “to be forced to deal hardly with you and
this sweet lady: whatever you be, King or poor Calabrian, you seem to love
your wife, so I would fain make you both comfortable. But the Auditor-
General (he that commands now) has charged me to keep you very strictly;
and since your wife insists on sharing your prison, she is to be served with
bread and water like yourself. I am heartily sorry for it, Sir, but I must do
my duty.”
Sebastian bowed in token of reply, for his emotion choaked him as he
gazed on the heavenly smile which shone through the tears of Aziek; that
smile said how little she regarded the pains and privations of the body—and
at that moment he loved her dearer than ever, for never had her unrivalled
attachment been so perfectly displayed.
Stephano withdrew, leaving the husband and the wife to seek consolation
in the possession of each other’s attachment.
On the fifth morning, Sebastian was surprised by the appearance of the
Auditor-General with his secretaries, who entering his cell, regarded him
some time with severe scrutiny. “I am come hither,” said he, “to ask you for
the first and last time, whether you persist in your imposture? if you abjure
your crime, and consent to make public confession of it before all men, I
am commissioned by our sovereign, Philip III. to promise you life and
liberty: but if you continue thus to maintain a falsehood, you will either be
left to linger out your days on bread and water, or perish at once by the
hands of the executioner. What is your reply?”
Sebastian turned on him a look of exceeding majesty: “I disclaim your
authority with that of your master, for I am his equal and his kinsman: let
him do with me as he will, I will still call God to witness that I am that self-
same Sebastian King of Portugal, who in the year 1578, passed into Africa
against the Moors; and the very same, who to augment the name and the
power of the Christians, put his life to the hazard, together with that of
twenty thousand brave men, whom his criminal obstinacy devoted to
slaughter. I am that unfortunate Prince, who for the punishment of his sins
lost the battle of Alcazar!—this is a truth which I may not deny without
endangering my immortal soul. Deal with me as you are commanded, I will
continue to utter the same words, in prison or at the stake.”
Sebastian turned from him as he concluded, and awed by his royal
manner, the auditor with his notaries (who had taken down the King’s
words in writing) departed without further speech.
Day after day now lingered by, and as they passed they cast a deeper
gloom over the prospects of Sebastian. The Count Lemos grew worse, and
Sossa (naturally of a harsh temper, and devoted to Philip) prohibited the
slightest mitigation of suffering to the unfortunate Sebastian.
Not for himself did Sebastian grieve, but for her whose tender heart and
delicate frame, were so ill suited to the rigours of their destiny. Yet alas! his
grief was vain and powerless.
Nearly two weeks had elapsed, when Stephano entered with a glad look,
to communicate the news of Count Lemos’s disorder having taken a
favourable turn; and to assure Sebastian that his friend the secretary (whose
mediation had procured to Kara Aziek the liberty of sharing her husband’s
fate) had promised to inform his master, of the peculiar severity with which
the alleged Calabrian was treated.
The secretary kept his promise. No sooner was Count Lemos in a
situation to investigate business, than he granted Don Sebastian permission
to appear before him, and for that purpose had him brought privately to his
house.
Neither time nor suffering could wholly deface the rare lineaments of
him, who might once have stood forth the model of manly beauty. Lemos
was not long of recognising in this interesting stranger the noble and heart-
winning Sebastian: he looked at him with sorrow and surprize; and having
questioned him on several matters known only to themselves, he
acknowledged himself convinced.
But the old nobleman was too well versed in the character of ambition,
to hope that Philip’s persecution arose from a real belief of imposture: he
justly thought that his august prisoner was secretly devoted to a lingering
death, and thus trusted to his keeping, from the apprehension, that if
brought either into Spain or Portugal, his escape would be productive of
more immediate danger, or his death exasperate the people into a revolt.
Lemos could only promise what he sincerely meant to perform, a strong
testimonial to the truth of Sebastian, and a consequent remonstrance with
his royal master: should that fail, he must content himself with watching
over the life of his prisoner, and yielding him all the comforts within his
power: to permit his escape, a nice sense of honor forbade.
“Whatever be the trust reposed in me, Sire!” he said, “if I accept it, I am
bound to hold it inviolate: and as my respect may sweeten your majesty’s
hard destiny, to refuse the charge of your person would be only to deliver
you up into the hands of a severer guardian.”
Too grateful for any amelioration of his fate, since that of Kara Aziek
was inseparable from it, Sebastian urged not a single argument against the
opinion of Count Lemos: he bestowed a warm eulogium upon his justice
and generosity, and accepted with gladness the offer he made him of future
protection.
By Lemos’s orders, the royal prisoners were removed into the best
chamber of the fortress, where Stephano and his sister were permitted to
wait on them. Books, musical instruments, and occasional walks in the
garden, under certain restrictions, now lightened their captivity: air and
better diet quickly restored some bloom to the cheek of Kara Aziek, and the
information (which she covertly obtained during the visit of Count Lemos)
that the Medina Sidonia family remained undisturbed, brought back some
peace to her mind. But anxiety for the ultimate end of their misfortunes,
devoured the inmost part of her heart, and like a canker-worm, preyed on
the source of life.
It was well for Sebastian that some innocent recreations enlivened his
captivity, since the prospect of ever being released, seemed daily less
probable. Philip’s answer to Lemos, had been in his usual strain of artful
moderation: afraid of exasperating that most respectable of his nobles, into
a revolt from his authority, and a public espousal of Don Sebastian’s
interests, he deemed it wise to tolerate him in dispensing those indulgencies
to the prisoner, which he boldly avowed his intention of always allowing;
while at the same time he peremptorily forbade the viceroy to write or to
speak to him in defence of an impostor. This title Philip scrupled not to give
him, in defiance of the Count’s testimony, being determined to resolve
every difficulty into the unreal solution of those days, absolute sorcery.
Several of the Portuguese, who had openly taken part with Sebastian,
were outlawed, and their properties confiscated: amongst them were Don
Christopher of Crato, and Juan De Castro.
Braganza’s high birth and vast influence alone saved him from feeling
the heaviest weight of Spanish resentment: policy taught Philip not to
exasperate the Portuguese too much, and Braganza was therefore spared.
But in the persons of his retainers he felt the malice of his rival:—Father
Sampayo was cast into the cells of the Inquisition on spiritual charges; and
had not Texere escaped into England, (where Sir Anthony Shirley for the
love he bore his master, granted him an honorable and safe asylum) he too
must have groaned in the same dismal prison.
Either by threats or bribes, the Spanish King had allured into his views,
nearly all of his courtiers that had been hardy enough to plead for a fair
scrutiny of the pretender. Rome had launched her lightnings and terrified
France again into silence: and in England, the disastrous fall of Essex, the
death of Elizabeth, and the succession of James, had changed its politics,
and rendered any expectation of support from that quarter a vain chimera.
To pass their lives in the castle Del Ovo, was therefore the last prospect
that remained to Sebastian and to his blameless wife. When our fate appears
inevitable, who is it that weakly continues to contend against it? Confiding
their daughter’s future happiness to Heaven, and to the Duchess Medina
Sidonia, they dried their tears, as they sometimes flowed, when thinking of
their eternal separation, and taught themselves to rejoice in her liberty.
Of the world they now thought only as of a scene on which they should
never more appear: they banished its hopes, its fears, its anxieties, and
submitting to the divine decree, made their world in each other’s hearts.
Those qualities which had never failed to attract and to attach every one
within their influence, still continued to win the affections of whatever
persons approached them. Stephano and Baptista privately confessed to
their friends, their admiration of the royal sufferers, and their firm belief of
Sebastian’s just claim on liberty and dominion: these confidential
discourses, spreading from confidant to confidant, at length diffused
throughout Naples so lively an interest in the supposed impostor, that Sossa
and others of Philip’s party became uneasy, and remonstrated against the
indulgence of Count Lemos.
While persisting in his generous line of conduct, the good Lemos was
seized by a return of his disorder, and in a very few days reduced to the
brink of the grave: his son, who was just arrived from the Spanish court,
and who came hotly zealous for the punishment of him, whom Philip
affected to consider a base-born Calabrian, was summoned to the deathbed
of his aged parent.
Count Lemos spoke of the prisoner: having listened patiently to the short
but violent reply of his son, Lemos raised himself on his pillow, and
addressing him with a solemn voice, said, “I am dying, my son! and the
words of a dying man may be trusted.—As I hope for mercy and pardon at
the judgment seat of Christ, I believe this man whom you call an impostor,
to be the true and lawful Sebastian King of Portugal: as such I charge you
(should my government devolve on you) treat him nobly; and let no worldly
honours tempt you to touch his life, or to connive at the violence of others.
Friends! you who surround and hear me at this awful moment, I charge you
all to testify what I have said, and to bear with it, my dying request to my
sovereign master Don Philip: I intreat him for his soul’s sake, to sift this
matter more closely.”
Exhausted by this exertion, Count Lemos stopt, and laid his head back
upon the pillow:—shortly after he breathed his last, and nothing remained
of the venerable old man, but a clay-cold corpse.
This event was a fatal blow to the comparatively happy state of
Sebastian and Kara Aziek: they were immediately remanded back to their
dungeon by Sossa; for Lemos feared Philip too much to obey his departed
father, and compromising with his conscience, by resigning the invidious
task into another’s hand, pretended that an excess of filial grief, made him
unfit to investigate so momentous a subject.
This severe treatment was followed by a visit from the stern auditor: he
came to demand a second time, the outraged King’s reply to his insulting
questions. Again Sebastian declared, that were he to live a thousand years,
and every hour of that long period to be employed in making the same
demand, he could not return any other answer than that he would live and
die professing his truth and his wrongs; that he appealed to a public trial in
his own dominions; that he protested against the injustice of his kinsman’s
proceedings, and would persist in doing so to his last breath.
“Your sentence is then pronounced,” returned Sossa, as he departed,
“your obstinacy condemns yourself: our illustrious and long-suffering
monarch has condemned you for life to the galleys.”
As the auditor disappeared, Sebastian fixed a fond but sad look on the
agitated features of Aziek. “Faint not, my beloved?” he said, “our appointed
trials must be bravely borne to the last—every species of oppression and
insult are to swell the cup of your Sebastian’s destiny; but remember the
bitterness of that mortal draught is short, in comparison with the eternal
spring, of which, through God’s grace I hope we shall drink together in
Heaven.”
Kara Aziek smiled with a breaking heart, and filled with admiration of
her husband’s magnanimity, earnestly prayed for strength to imitate so
noble an example.
On the day which removed Sebastian to this new scene of misery, he was
led from his prison to be conveyed to the gallies. Lemos and Sossa believed
that to shew this compassionated sufferer to the expecting crowd under
degrading circumstances, would be a surer antidote to their respect, than if
they beheld him brought to public execution: they had therefore decreed
that he should be led through the streets of Naples to the port, mounted on
the most ignoble of animals, and followed by his faithful Aziek, in the
meanest attire.
At the gate of the castle, he beheld multitudes of soldiers and spectators,
and a herald holding the ass upon which he was to mount:[A] his
countenance was unchanged: he placed himself on the lowly animal with a
serene and majestic aspect that might have become a throne; it ennobled his
sorry garments, and touched every beholder with respect and pity.
[A] This incident is an historical fact.
His eyes, (brightening as he moved) were fixed upon Heaven: it seemed
as if in this triumph over human weakness and human passion, he felt the
blessed earnest of eternal reward.
As Kara Aziek hastened to follow the slow progress of Sebastian, some
unfeeling wretches scoffingly bade her behold the King her husband, and
admire the splendor of his array; she flashed on her insulters a glance of
honourable indignation, for in her heart was love and veneration united for
him they contemned. No fear, no shame could find entrance there: love
raised to enthusiasm by grief and admiration, irradiated her features, and
gave its former bright flush to her burning cheek: a single black garment
wrapt her somewhat wasted figure; her head and feet were bare, by orders
of the merciless Sossa; but those delicate feet rending their tender surface
against the sharp pavement of the streets, and that hair which fell
dishevelled in all its beauty around her, excited only the more compassion.
Beauty is the most touching orator; and the loveliness of Kara Aziek
heightened the effect produced by her devoted attachment to the husband
she followed.
As they moved along, preceded by a herald, proclaiming the offence and
the sentence awarded to Sebastian, the murmurs which at first rose among
the crowd, gradually died away, till an awful and unbroken silence
universally prevailed. The people looked on each other with sorrow and
amazement; while Sebastian now and then removing his eyes from Heaven,
looked round upon the spectators with pardon and pity for their sin of
consenting to so black an act. At times, when the herald called aloud, “this
man whom traitors assert to be the King of Portugal, &c.” Sebastian would
interrupt him in a loud voice, exclaiming, “and so I am,”—then resuming
his calm attitude, proceed in dignified silence.
Arrived at his place of destination, he turned to bid farewel to some of
his humblest friends, whom he recognized amongst the crowd. “Friends!”
he said, “ere you lose sight of me for the last time perhaps, bear witness that
I testify to the truth of my own assertion: I am Sebastian King of Portugal;
this matchless woman my lawful and beloved wife. I submit humbly to the
will of God, not basely to the oppression of man: my body I account for
nothing; and upon that only, may Philip heap indignity and pain; my soul, is
above his reach.
“From such of you as have suffered, or may suffer loss for my sake, I
crave pardon and pity; the most grievous of my sorrow, is the consciousness
of having caused sorrow to others: Heaven will reward you, since the
unfortunate Sebastian has no longer any thing to bestow but his poor
thanks.”
He had scarce spoken, when the sound of weeping was heard, and a
woman pressing forward, threw herself on the earth before him and Kara
Aziek: it was Paula the widow of Gaspar. At sight of her, Sebastian turned
pale from excess of emotion, and tears gushed from his eyes. “My poor
Gaspar!” he exclaimed, “I regretted thee when I should have rejoiced! hadst
thou lived to see this woeful day!”
Sebastian stopt, for Kara Aziek’s long-stifled grief, now burst forth with
such passionate violence at the prospect of separation from him (for she
was denied the consolation of sharing his destiny) that his fortitude began to
faint, and his limbs shook as he sought to support her.
Meanwhile Paula was calling on the people around, to witness, that since
she now saw the person whom they had taken her to Venice to swear was
Marco Cattizone, she denied his being so; that she recognized in him the
former master of her deceased husband, and not that dear husband himself.
She wept the memory of Gaspar with unfeigned sorrow, which encouraged
Sebastian to require her care of his Aziek, for whom no better habitation
offered an asylum than the humble one of Baptista and Stephano.
These good people had promised to receive and to comfort Kara Aziek
at the fatal moment in which she must be severed from the partner of her
life. Baptista engaged to effect occasional interviews between the wife and
husband, through her influence over a young man in the galley to which
Sebastian was doomed; and it was from this promise that Kara Aziek gained
strength to live through the wretched scene in which she was now
performing.
Amidst tears, embraces, lamentations, and exhortations, she was torn
from the arms of Sebastian: he was hurried into the galley, and she led
almost lifeless away to the lodging of Baptista.
Associated with slaves and malefactors, behold the once imperious and
fiery King of Portugal submitting to his destiny with a resignation that gave
a dignity to humiliation: he arraigned not Providence, for he remembered
his past bigotry; and though the humanity of his nature had prevented him
from carrying his zeal into absolute persecution, he ventured not to say how
far that zeal might eventually have transported him; devising it just that he
should find his present punishment from that disposition in others, which he
had cherished in himself.
Even the most merciless of his oppressors in Naples, dared not outrage
that dignity which awed them, by insisting on his labouring like a criminal
at the oar: he was condemned to the galley merely as to a more public
prison. Philip thus making a show of believing that the more he was seen,
the less he would be credited; but in reality hoping that one of those
malignant fevers common to the galleys, would soon send him to another
world.
Stephano kept his word, and procured more than one meeting between
the royal sufferers. Paula undertook to transmit an account of the King’s
situation to his kinsman Braganza, by whom her infant was now protected.
She herself was thus far on her way to rejoin the Duchess of Medina
Sidonia in Spain, whither Paula found her gone to take leave of a dying
friend. Paula used all her rhetoric to persuade Kara Aziek to accompany
her: but not even the temptation of beholding Blanche again, could swerve
the conjugal love of Aziek: she was determined to follow the fate of her
husband, wherever it might lead; and as the Duchess held in her possession
the little remnant of their wealth, and might transmit it through Paula, Aziek
resolved to avail herself of Stephano’s protection, and dwell at least in the
vicinity of the galleys.
Charged with letters to their beloved child, and to the Duchess, in which
the anxious parents besought all efforts for their own happiness to cease,
and nothing be attempted but for that of Blanche, Paula departed from
Naples.
No sooner was the injured King of Portugal placed in a situation which
exposed him to all eyes, than crowds flocked to see and to converse with
him. Every day, every hour, produced fresh testimonies to his truth: and had
not religious prejudices enfeebled their compassion, and lowered their
respect, the Neapolitans would have joined the loyal Portuguese in rescuing
him by force from the galleys.
This universal discontent so far alarmed the new Count Lemos, that he
dispatched a messenger to Madrid, with a detail of what he feared: the
consequence of his dispatch, was an order for the galleys to quit the
Mediterranean and come down to the western coast of Spain.
Not even this change, could divide Kara Aziek from Sebastian: she
followed him in a little vessel bound for the same port, accompanied by
Baptista, whom kindness and fidelity had endeared to her, and rendered her
chief solace.
The galleys were commanded to ride in the bay of St. Lucar: and at St.
Lucar Kara Aziek took up her abode.
Unknown and unnoticed, she depended solely on the humanity of
Baptista’s lover, for distant interviews with her husband.
The saddest period of Kara Aziek’s life was now present: she lived
forlorn of every comfort except only the humble attentions of her servant,
and the occasional sight of him from whom she once fondly hoped nothing
less than death would ever have parted her. She beheld this object of her
heart’s idolatry, loaded with chains, and condemned to the vilest of human
stations: her imagination pictured the closing scene of this dismal tragedy,
and presented him at the stake, or on the block. Her amiable daughter was
now far away, and too probably the eyes of each fond parent would close
for ever without beholding her again. Their private friends were dead, or
dispersed; their more potent ones, the sovereigns of Europe, changed by
circumstances, or rendered powerless from necessities of their own. All
around was dark and dreary; and wherever she looked, still the same black
horizon shut in her fate.
Where is the spirit that can resist calamities so heavy? Religion may
enable us to curb complaint, to submit with humility and a thorough
conviction that he who ordains, is all-wise, and all good; but not even
religion can benumb “the nerve whence agony is born:” The heart may
break while it yields.
Kara Aziek felt hers to be fast decaying: sorrow wasted her bodily
strength, and with it her mental energy. A deep sadness was fixed upon her
countenance, and heavy and continual sighs (of which she was herself
unconscious) told the attached Baptista, that her suffering mistress was
hastening to the repose of Heaven.
At this period, Baptista unexpectedly heard that the family of Medina
Sidonia had a residence in the neighbourhood of St. Lucar, and were
coming to visit it: she imparted this to Kara Aziek, believing Paula likely to
be in the Duchess’s suite, and knowing of no other interesting object; this
intelligence lifted up the soul of the fond mother; she hazarded a letter to
the Duchess through the medium of Paula, and remained with trembling
eagerness anticipating an answer.
CHAP. VII.
Some days had passed in anxious expectation, when in the dusk of
evening, a man in a domestic’s habit appeared at the obscure abode of
Baptista, and announcing himself sent by the Duchess Medina, urged
admittance to Kara Aziek. She received him alone.
Having cautiously closed the door, the stranger threw off his cloak and
hat, and Kara Aziek saw at her feet a young and handsome man, on whose
intelligent countenance nature’s hand had stamped truth and goodness.
“It is the Queen of Portugal to whom I bend my knee?” said the
animated youth, with a look that demanded if he were right. Kara Aziek
answered with ready tears, “Alas! it is the most desolate of women—the
wife of him whom would to God I could say was not born to a throne but
you come from the Duchess of Medina Sidonia—know you her young
companion—Blanche?”
A graceful disorder appeared on the face of the stranger, as with
deepened colour and a fluttering voice, he replied that she was even then
near St. Lucar. Kara Aziek clasped her hands together in an ecstasy of
gratitude: but the emotion of the stranger did not escape her; and his menial
habit so ill suited to the elegance of his air, made her attach an agitating
meaning to the emotion he betrayed.
“I do not see a domestic of the Duchess Sidonia’s?”
“No, Madam—you behold her son.”
Don Hyppolito now put into her hands a letter from his mother, which
Kara Aziek eagerly read. It informed her that Hyppolito was zealous in the
cause of Don Sebastian, though wholly ignorant of their lovely ward’s
connexion with his fate; that the Duchess was eager to afford the parents a
sight of their daughter; and that the Duke having been fortunately
importuned by all the Spanish grandees around, to visit the newly arrived
galley, in which the King of Portugal was confined, in order to disprove the
impostor from his personal knowledge of the true Sebastian, he had seized
the opportunity of yielding to his own earnest desire, and was come openly
to St. Lucar for the avowed purpose.
Desirous of bringing the mother and child to an immediate meeting, the
good Duchess had sent her son (disguised thus to prevent observation) with
orders to attend Kara Aziek to their dwelling near the town. Hyppolito (thus
ended the letter) has no suspicion of the relationship we so religiously
conceal, it will depend on yourself and the royal Sebastian, whether he may
ever be so greatly trusted.
But Hyppolito scarcely needed to be now informed of the momentous
secret; the likeness to Blanche, which a lover’s eye directly discovered,
together with a recollection of Blanche’s distracted sympathy with the
sufferings of the King and Queen of Portugal, and now the wild joy of Kara
Aziek, all united to shew him the fact. Trouble and apprehension succeeded
to his lively enthusiasm; and a multitude of strange pangs seized his young
heart, as he prepared to lead forth the trembling mother.
During their hasty, and rather long walk, no words were exchanged
between them; Hyppolito stopped under the high wall of a garden, and
opening a small door concealed by trees, conducted Kara Aziek in. The
next moment brought them to a pavilion, where he would have left his
companion to enter alone, (so his mother had instructed him) but detaining
him by the arm, she exclaimed in a low voice, “O no—leave me not—I owe
you the reward of seeing how happy you have made me.”
Hyppolito caught at the permission; he pushed open the door, and the
next moment beheld the mother and the daughter senseless in each other’s
arms.
Joy and grief so blended, were too powerful for their hearts: nature sunk
under such a meeting, and it was long ere the Duchess and her son
succeeded in restoring them to life.
Sorrowful happiness was that which the young Hyppolito now
witnessed; no sounds, save those of weeping and sighing, were heard
through the apartment. Though the mother and daughter fondly embraced,
fondly gazed on each other, their hearts were full of Sebastian, and
incapable of real joy.
The tears of Blanche flowed with redoubled impetuosity whenever she
looked on her mother. What a change did she see in that face and that
figure! as her eyes wildly noted the ravages made there by sickness and
sorrow, cold chills crept through her veins; she felt that a moment was
approaching in which she would require the consolation of some object
equally dear, and her eyes then sought those of him to whom her innocent
heart unconsciously trusted for all its future comfort. Hyppolito’s soul
speaking from his face, answered the supplication of hers: he advanced, and
joining the hands of Kara Aziek and her daughter, in one of his, he pressed
them with trembling lips, while a tear fell from his cheek upon the hand of
Blanche. Kara Aziek smiled benignly, and returned the affectionate
pressure.
To proclaim her maternal claim on the love of Blanche, seemed needless;
but Kara Aziek gratified the Duchess by requesting her to place that
confidence in the young Hyppolito. After a hasty explanation, the Duchess
ventured to offer some incitement to hope of better days, lamented the
circumscribed power of her husband, but assured Kara Aziek, that after he
had seen Don Sebastian, by the desire and in the society of those noblemen
who had urged him to the interview, and had convinced himself, by ocular
proof of his identity, he would boldly publish the truth at all hazards. The
Duke was now absent at the Governor of St. Lucar’s, but the morrow was
pitched on for his visit to the galleys.
This information infused a faint hope through the bosom of his wife; she
recovered by degrees from the excess of her first emotion, and remained till
night was far advanced, tasting a sad pleasure in noticing the ardent and
respectful passion which now blazed out, now receded from the fine eyes of
Hyppolito, and fitfully coloured the cheeks of the bashful Blanche.
This love unknown to themselves, even while for ever felt, was not
unmarked by the Duchess; and her looks had already interrogated those of
Kara Aziek, with a sort of pleadingness for her son’s happiness, which gave
the most solid satisfaction to the anxious mother.
Kara Aziek returned from this interview with a placidity long unknown
to her; and Baptista, who merely guessed that she had been visiting Paula,
made no inquiries, contented to observe that her mistress was really less
dejected than usual.
But violent emotions, whether sad or exhilarating, are equally dangerous
to a weakened frame: Kara Aziek was unable to rise from her humble couch
on the ensuing morning, when Hyppolito came to inform her he was going
with his father to recognize Don Sebastian.
Her death-like paleness (over which a smile of grateful regard cast the
brightness of immortal beauty) touched the romantic heart of Hyppolito,
and as he earnestly regarded the lovely wreck before him, love and pity
inspired him with the determination of attempting something to smooth at
least, her departing hour.
Having received a tender message for Don Sebastian, he hurried back to
Blanche, whose duteous love did not wait for the disclosure of his wish ere
it prompted her to exclaim. “So ill! so desolate!—O, Hyppolito, since my
dear mother may not dwell here unsuspected, I will go to her habited less
gaudily: confined to her sick chamber alone, in such an obscure quarter of
the city, who will know the adopted child of the Duchess Medina Sidonia?
—Some excuse may be invented for my absence, to prevent the curiosity of
domestics—Ah! if she were to be torn from me, without my having the
consolation of——”
Tears choaked her utterance, and covering her face, she remained
abandoned to sorrow, while Hyppolito was urging the Duchess to sanction
their pious project.
What mother could refuse such pleadings? Blanche was allowed to
follow the impulse of filial tenderness; her dress was secretly exchanged for
one of Paula’s, and gliding unseen through the garden, Hyppolito conducted
her out of the private door, and led her safely to the arms of her expecting
mother.
The ardent young man had not time to do more than kiss the hand of
Kara Aziek, ere he ran off to join his father, and the rest of the grandees.
Accompanied by his wife, the Duke of Sidonia proceeded to the shore;
his aspect was grave and thoughtful; for he was reflecting on the wondrous
vicissitudes of our mortal life. That unfortunate Prince, upon whom all men
might now gaze unchecked, all tongues move in reviling, was that same
Sebastian whom Medina had last beheld, surrounded by power and majesty.
It was that King whom Medina had himself served twenty years before,
with submissive awe; whom he had feasted and entertained with tilt, and
tournament, and ball, while he waited at Cadiz for the troops of Philip II.
These reflections occupied the Duke till his company reached and mounted
the chief galley.
Hyppolito was the first to spring on deck: he looked eagerly round, and
immediately singled out the august object of his search. Removed from the
other slaves, in a lonely quarter of the ship, he saw a man seated, with his
arms folded, and his head bent towards the ground; his single garment was
coarse and dark; his head and limbs were without covering; but the large
and noble proportions of those once powerful limbs, and the majestic air of
that head, denoted him to be the King of Portugal.
Hyppolito hastily advanced, and his quick breathing stirred the attention
of Sebastian; he looked up, his eyes met those of Hyppolito, who felt them
enter into his soul. By a sudden impulse, the young man half bent his knee;
surprise and inquiry illuminated the countenance he was observing.
Sebastian slowly arose, and as he did so, his youthful companion heard the
clank of chains.
Such an expression of shame and indignation banished the air of
veneration with which Hyppolito was looking at him, that Sebastian
understood what passed in his mind. “Young man,” said he, “blush not for
me—blush for my oppressors, and my coward friends!—deserved
punishment is disgrace—but unmerited oppression, if nobly borne, is
glory!”—He moved away as he concluded, leaving Hyppolito gazing after
his kingly step, and yet—commanding figure.
The vessel was soon crowded with illustrious visitants from the yacht of
Medina Sidonia; the captain of the galley understanding their errand,
shewed the Duchess and her company to a wider part of the deck, and sent
to inform Sebastian that they entreated to see him. He turned back with the
captain, and calmly advanced into the circle formed by his examiners.
The Duchess who had last met him at Villa Rosolia, under such different
circumstances, almost uttered a cry of melancholy welcome: her company
burst forth into remarks and questionings: the Duke remained on one spot,
steadily eyeing the figure before him.
So long was his scrutiny, that some of the group impatiently demanded
whether the man they saw, were not really an impostor. Medina suffered
them to importune him for an answer, and at length seriously replied, “Am I
to speak the truth my lords?—I declare then, that in the voice and mien of
this stranger, I recognize the very voice and mien of the King of Portugal.
The alteration I find in his face and figure, is only such as twenty years of
suffering might be expected to produce.”
“What then, you believe he is Don Sebastian? You assert it?”
“I assert nothing: persons and voices may resemble; but in events we
cannot be mistaken. If this be the King, whom I entertained at Cadiz, ere his
expedition to Africa, he will be able to point out to me amongst some
armour which I have had brought hither, the present he made to me at that
period.”
“I gave thee a sword, Sidonia!” said Sebastian, “and I think I should
remember it again.”
The surrounding nobles, with dismay and surprise, followed the Duke to
the stern of the galley, where some attendants had just arrived with a heap
of swords, spurs, curious pistols, and daggers. The Duke silently pointed
out to them all, the weapon given him by the King, which being less costly
than any of the others, was the least likely to be guessed at as a royal gift.
Sebastian, who had remained exchanging looks of interesting meaning
with the protectress of his daughter, courteously went to meet the returning
party; an old servant displayed the armour: Hyppolito bent anxiously
forward, fearful, that if the King’s memory failed of retaining such a trifle,
they who chose to cavil at this truth, might seize so plausible a pretext, and
pronounce his father deceived.
But, at the first glance, Sebastian recognized his own plain sword, and
drew it from beneath a heap of others. “With this sword did I make thee a
knight of Avis!” he said, sorrowfully, “O sad remembrance! for what a train
of bitter recollections is in its train!”
“Now, my lords, what say you?” exclaimed the indiscreet Hyppolito,
—“should you not bow your knee and acknowledge the royal kinsman of
our sovereign Philip of Spain, and should we not all join in bringing this
convincing proof to his abused ear?”
Most of the nobles, who well knew that the ear of Philip was wilfully
stopt, fell back, murmuring “Sorcery or accident,” while others expressed
their conviction, but lamented their want of influence. The captain of the
galley stood with an air of sincere remorse, which did not escape Hyppolito.
The old servant holding the armour, having carefully examined the
lineaments of the King, added his testimony to that of his master. The deck
of the galley became for a while a scene of confusion and strong emotion.
Sebastian alone, was little moved; he was no longer to be deceived by vain
hopes; he knew that all those people would go home convinced of his truth,
pitying his misfortunes, and earnest in wishing them at an end; but that in a
short time their wonder and their concern would cease; he would be
forgotten, and left to his fate.

Geostatistical Functional Data Analysis Wiley Series in Probability and Statistics 1St Edition Jorge Mateu Editor Full Chapter

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Geostatistical Functional Data Analysis Wiley Series in Probability and Statistics 1St Edition Jorge Mateu Editor Full Chapter

Uploaded by

Copyright:

Available Formats

Geostatistical Functional Data Analysis

(Wiley Series in Probability and

Limit of Liability/Disclaimer of Warranty

Library of Congress Cataloging-in-Publication Data

Names: Mateu, Jorge, editor. | Giraldo, Ramón, editor.

Cover Design: Wiley

Set in 9.5/12.5pt STIXTwoText by Straive, Chennai, India

List of Contributors xiii

1 Introduction to Geostatistical Functional Data Analysis 1

Part I Mathematical and Statistical Foundations 27

2 Mathematical Foundations of Functional Kriging in Hilbert

2.3.4 An Application to Nonstationary Prediction of Temperatures

3 Universal, Residual, and External Drift Functional

4 Extending Functional Kriging When Data Are Multivariate

4.6.1 Multivariate FPCA 91

5 Geostatistical Analysis in Bayes Spaces: Probability

6 Spatial Functional Data Analysis for Probability Density

6.4 Conclusion 149

Part II Statistical Techniques for Spatially Correlated

7 Clustering Spatial Functional Data 157

8 Nonparametric Statistical Analysis of Spatially Distributed

8.A.2.3 Proof of Lemma A.4 196

9 A Nonparametric Algorithm for Spatially Dependent

10 Nonparametric Inference for Spatiotemporal Data Based on

11 Modeling Spatially Dependent Functional Data by Spatial

12 Quasi-maximum Likelihood Estimators for Functional Linear

13 Spatial Prediction and Optimal Sampling for Multivariate

Part III Spatio–Temporal Functional Data 351

14 Spatio–temporal Functional Data Analysis 353

15 A Comparison of Spatiotemporal and Functional Kriging

16 From Spatiotemporal Smoothing to Functional Spatial

Ana M. Aguilera María del Carmen Aguilera Morillo

Mohamed-Salem Ahmed Sophie Dabo-Niang

Mara S. Bernardi Maria Durban

Laurence Broze Zied Gharbi

Ramón Giraldo Dae-Jin Lee

Alberto Guadagnini Claude Manté

Alessia Pini Johan Strandberg

Cristian Preda Camille Ternynck

November 2020 Sir Bernard Silverman

Introduction to Geostatistical Functional Data Analysis

1.1 Spatial Statistics

defensible inferences on processes and parameters of interest. These modeling

consideration is that very often, simple topography makes a stationary assumption

1.2 Spatial Geostatistics

1.2.1 Regionalized Variables

1.2.2 Random Functions

The function 𝛾(si , sj ) is called semivariogram.

1.2.3 Stationarity and Intrinsic Hypothesis

intrinsically stationary random function or random function of stationary incre-

E (Z(s)) = 𝜇(s) = 𝜇. (1.4)

C (Z(s), Z(s + h)) = C(h), ∀s ∈ D and h. (1.5)

As the covariance function C(h) of a second-order Stationary, r.f. is only a func-

V(Z(s)) = C(𝟎) = 𝜎 2 . (1.6)

dependence displayed by the phenomenon, as they verify the following mutual

If a r.f. is second-order stationary, then it is also intrinsically stationary. How-

1.3 Spatiotemporal Geostatistics

1.3.1 Relevant Spatiotemporal Concepts

Separability is a particular case of complete symmetry and, as such, any test

Whenever it is possible to define the covariance function and the variogram,

for any n ∈ ℕ, (si , ti ) ∈ ℝd × ℝ, and ai ∈ ℝ, i = 1, … , n. The condition (1.20) is suf-

Schoenberg [10] proved the following theorem characterizing the spatiotem-

● 𝛾((si , ti ), (sj , tj )) is a semivariogram on ℝd × ℝ.